Re:Harvard researchers find AI coding agents shifted the bot...
by judahmeek on Tuesday November 17, 2026 @04:16PM
(#50538304)
I've submitted the original paper (https://fion.ac/jellyfish.pdf), but others already have to little engagement, likely due to the paper's terrible title.
Re:Harvard researchers find AI coding agents shifted the bot...
by u1hcw9nx on Tuesday November 17, 2026 @04:20PM
(#50538349)
This again increases the demand for good programmers.
Write only programmers are dime on dozen. Those who can read code written by others (and themselves) and analyze it are harder to find.
Re:Harvard researchers find AI coding agents shifted the bot...
by diego_moita on Tuesday November 17, 2026 @07:22PM
(#50540233)
No, it is quite the opposite. In the scale that AI produces code, it isn't feasible to have humans reviewing it. And, frankly, reviewing code sucks big time.
The solution is to find ways to automate the process of code approval: more thorough tests and reviews, both made by AI.
Re:Harvard researchers find AI coding agents shifted the bot...
by lo_zamoyski on Tuesday November 17, 2026 @07:29PM
(#50540272)
If reviewing code "sucks big time", then I don't know what to tell you. Reading code is more important than writing it. LLMs can help out with reviews, but it cannot replace human judgement.
Re:Harvard researchers find AI coding agents shifted the bot...
by lo_zamoyski on Tuesday November 17, 2026 @07:27PM
(#50540261)
You would hope, though we'll have to see how management actually sifts the chaff from the wheat in practice.
Code reviews have been notoriously neglected. In the most charitable cases, this was the result of being too bogged down with work or some kind of implementation minutia or time-consuming plumbing to have enough time to actually review code with enough care. That and you aren't necessarily rewarded for code reviews by mediocre companies, so there are perverse economic incentives. Which is disconcerting, because it creates a compulsion to produce without understanding, and understanding is more valuable than artifact.
Reading code is indeed more difficult, because the code's intent, composition, and architecture must be pieced together and inferred with the help of context (formalized languages are infamously bad at communicating intent and most code in the world is full of mixed and half-baked abstractions that further serve to obfuscate that intent, and intent is the measure). So a good code review of anything substantial involves thoughtful effort.
Now that LLMs can catch all sorts of trivial anti-patterns, the need for good judgement and reasoning is hopefully made more conspicuous, because as we know, vibe coders will continue to brush off code reviews, and mediocre programmers will continue to perform them poorly.
Re:Harvard researchers find AI coding agents shifted the bot...
by timr on Tuesday November 17, 2026 @09:30PM
(#50541113)
> Those who can read code written by others (and themselves) and analyze it are harder to find.
It helps when your industry-wide hiring process isn't an institutionalized speed test of ridiculous code golf problems that is memorized from a ridiculous book by every new generation of programmers.
It's almost like we should have been looking for people who can read code all along. But that would be crazy.
Re:Harvard researchers find AI coding agents shifted the bot...
by lelanthran on Wednesday November 18, 2026 @05:36AM
(#50544523)
> Those who can read code written by others (and themselves) and analyze it are harder to find.
They may be harder to find, but what good are they when no one is reading code in-depth anymore?
I mean, in 2022, coders who can code without syntax highlighting and autocomplete were hard to find, but no one cared anyway because that was not a skill that was needed.
And, a few years from now, coders who can code without AI will likewise be hard to find, but no one will care because no one wants coders who can write code without AI.
Re:Harvard researchers find AI coding agents shifted the bot...
by gzell on Tuesday November 17, 2026 @04:56PM
(#50538791)
From real-world experience, I have to disagree. It always has been and continues to be a faulty, up-front specification problem. The tail don't wag the dog.
Re:Harvard researchers find AI coding agents shifted the bot...
by hibikir on Tuesday November 17, 2026 @09:10PM
(#50541000)
We were there many years ago too though: I had to take over a rather complicated system circa 2012, which had been designed to be built by a team of 20 or so, for timelines, but they had little effort on integration. This meant that a whole lot of guards were written all over the place for situations that, in practice, never occurred. To try to get this to production in a timely fashion, I ended up putting metrics per guard, along with metrics for what was caught by broad error checks near the base of the program. Basically all the defensive programming that had been built was pointless, while I had to write a whole lot of cases that were being dropped, and the people that designed things up front never considered.
AI error handling in code review is a lot like said defensive programming in a vacuum: It'll take a lot of code and a lot of energy, but it's unlikely to be important. Even when you are designing something that gets executed hundreds of millions of times a minute, over-review is easy. You pay for it when letting errors in the real world are unacceptable, but it's pretty inefficient. And that's without getting into features that just don't get used, so the entire thing should not have been built in the first place.
Re:Harvard researchers find AI coding agents shifted the bot...
by mathisfun123 on Tuesday November 17, 2026 @06:23PM
(#50539648)
> The tail don't wag the dog
If I increased the output of my pump jack and identified the refinery as the new bottleneck would you say the same pithy thing? NB a single pump jack can pump different grades of crude oil at the same time.
Re:Harvard researchers find AI coding agents shifted the bot...
by skydhash on Tuesday November 17, 2026 @07:36PM
(#50540318)
I’m not avoiding. But you have to state your assumption that everything before the pipeline is working well and it’s just the pipeline that is struggling.
Then your analogy would break down as we all know that the PR machine is a slop producer.
Re:Harvard researchers find AI coding agents shifted the bot...
by donkey_brains on Tuesday November 17, 2026 @07:09PM
(#50540116)
Here’s the thing: a human engineer is embedded in their company. They sit in meetings, they know whose opinions are more important and whose they can disregard. They know the history of how people at the company like to do things. They use all of this context to feed more precise specifications into a coding agent, and as a result, they get an output that better suits the environment of their company. Ergo, they spend less time reviewing agent output which was built to solve the wrong problems in the wrong way, and which ultimately must be discarded or significantly rewritten.
Re:Harvard researchers find AI coding agents shifted the bot...
by skeledrew on Tuesday November 17, 2026 @10:18PM
(#50541436)
No such thing as a perfect spec, but make a very good one, have a few agents also review there implementation against it, and then actually test the product manually. Gaps and bugs will almost always show during manual testing; skimp on it at your own risk.
Re:Harvard researchers find AI coding agents shifted the bot...
by wilkystyle on Tuesday November 17, 2026 @06:46PM
(#50539888)
With all respect to your personal experience, I could not possibly disagree more strongly with this.
In my own real world experience, the bottleneck has never been crafting the perfect specification up front. In fact, I have never, not even once, seen the initial specification for any real world software project of any mentionable size or scale survive unscathed to delivery. In reality, it is much like the old wartime maxim. "No Plan Survives First Contact With the Enemy".
The real world is messy and unpredictable. Every single software project of significance that I have been a part of has always involved testing, validating, and reworking the assumptions in the initial design and spec, refining them to arrive at the final product.
Furthermore, this iterative process (pre-AI) used to involve one or more humans following the development story and understanding what design changes and corresponding trade-offs were happening along the way (and why they were being made). By the time there was a finished deliverable ready for review, the person involved was privy to the story of why the development shaped up the way it did. They were able to account for the big picture, adapt accordingly along the way, and be able to explain why things ended up the way they did.
IME, with fully-agentic PRs, AI tends to view the specification as gospel and to view every point of friction between the specification and the real world as something to be worked around or minimized. Where a real human would pause, zoom out, and consider the system and bigger picture, considering what design changes might be needed, the AI will assume that its job is to make the specification a reality at all costs.
You can of course mitigate this by staying more closely involved with the work as it's being done, but this is of course slower than just throwing a spec over the wall and letting an agent run with it. It seems that people are far too eager to optimize for raw speed, and not nearly concerned enough with correctness or human understanding.
A sister comment mentioned a plethora of PRs that don't solve the right problem. If you are swamped with PRs solving the wrong problem, the root cause is not that you didn't foresee every future possibility or undiscovered devil in the details in order to create the right spec, the problem is that you thought you could come up with the right prompt at the outset and that an agent living in a clean digital world could make the journey all the way to a product fit for the messy real world on its own.
Re:Harvard researchers find AI coding agents shifted the bot...
by blurbleblurble on Tuesday November 17, 2026 @08:14PM
(#50540618)
I don't think they're talking about "the perfect specification" they're talking about correctly describing what the program should do at all, formally or informally.
Re:Harvard researchers find AI coding agents shifted the bot...
by aleqs on Tuesday November 17, 2026 @06:25PM
(#50539670)
And just because harvard researchers say something doesn't make it true either. Especially when it's like 6 months late and reviews are probably not the bottleneck anymore - testing, ci, parallelization are (imo).
Re:Harvard researchers find AI coding agents shifted the bot...
by nix-zarathustra on Wednesday November 18, 2026 @03:18AM
(#50543355)
Put it this way, AI's ass isn't on the line if it makes a mistake. The incentives push people to understand and own the code since they are ultimately responsible for any changes they make, with or without AI.
Re:Harvard researchers find AI coding agents shifted the bot...
by wj on Tuesday November 17, 2026 @09:38PM
(#50541157)
We are using AI to do the first pass at code reviews but they still are a bottleneck. The number of requests has exploded as has the amount of code. More is happening at all stages of the development cycle.
Part of a LinkedIn post I wrote in July:
--
A development cycle of a feature is roughly:
1. Request
2. Review
3. Requirements
4. Architecture
5. Development
6. Code Review
7. UAT
8. Deployment
AI has reduced the length of step 5 (development) and step 6 (code review). Code is a prompt away.
AI has increased top-of-funnel (1) volume. While each development step is shorter on average it is resulting in more instances of every other step. Throughput increases and the bottleneck moves one step further down the cycle. The shape of the work is changing from writing code to reviewing code.
Re:Harvard researchers find AI coding agents shifted the bot...
by jacobgorm on Tuesday November 17, 2026 @06:20PM
(#50539629)
The problem is that when nobody writes code or feels ownership over it anymore, nobody will be qualified to or really interested in doing the code reviews. The way forward could involve dropping code reviews altogether. You could argue that the useful pingpong between coder and reviewers has been replaced by the similar pingpong between the prompter and the agent.
Re:Harvard researchers find AI coding agents shifted the bot...
by prmoustache on Tuesday November 17, 2026 @06:30PM
(#50539720)
In my company, people found out that if they use an agent that commits and do the pull/merge request, they can approve it themselves and bypass a second set of eyes.
Since they are the one doing the prompting, they quickly got lazy of actually reviewing the output from the LLM which means that nobody is reviewing anything.
It hasn't happened yet but I guess that is how you end up with major outages.
Re:Harvard researchers find AI coding agents shifted the bot...
by taurath on Tuesday November 17, 2026 @06:53PM
(#50539955)
This is everywhere to my knowledge, and engineers are burning out left and right with stress because of how little anyone understands the code, and how much of a mountain of nobody knowing anything they have to handle. We are turning our critical infrastructure mechanisms into stochastic grey goo.
I find the engineers currently thriving are the ones who never gave a fuck about anything but their own advancement and are happy to drop weeks of work on others and take the credit.
Re:Harvard researchers find AI coding agents shifted the bot...
by guhidalg on Tuesday November 17, 2026 @08:11PM
(#50540603)
Understanding is overrated, application code (user-space) is anyway just a way to make money. If you're making more money over time, you're doing it right. We're not writing kernel code or safety critical code, we can let AI do the heavy lifting of putting characters into a file.
Re:Harvard researchers find AI coding agents shifted the bot...
by tcoff91 on Tuesday November 17, 2026 @08:27PM
(#50540706)
There are plenty of apps that while not safety critical, are financially critical to people's lives. When you're working on an app that processes billions of dollars in payments per month you can't just yolo it.
Re:Harvard researchers find AI coding agents shifted the bot...
by Jensson on Tuesday November 17, 2026 @11:56PM
(#50542050)
How? If they aren't at your place and are at his place, then he is right and you are wrong. You can't confirm its not done anywhere, but you can confirm its done somewhere.
Re:Harvard researchers find AI coding agents shifted the bot...
by kergonath on Wednesday November 18, 2026 @02:55AM
(#50543139)
> We're not writing kernel code or safety critical code
Well, you’re not until you realise that your AI has done something stupid and then you’re part of an exploit. Or exfiltrates used data to an unsecured cloud, or something equally smart.
Re:Harvard researchers find AI coding agents shifted the bot...
by Buttons840 on Tuesday November 17, 2026 @10:31PM
(#50541508)
Also, people who cause problems and then are in a personal situation where they can hop in and enthusiastically help fix the issue always look better than people who just don't cause problems.
Re:Harvard researchers find AI coding agents shifted the bot...
by Buttons840 on Tuesday November 17, 2026 @11:25PM
(#50541860)
> how little anyone understands the code
Everyone's doing code reviews--well, not even code reviews; just PR reviews--so how are they supposed to gain an understanding of the system.
I review a PR, I see that one new endpoint was added to the API. It looks correct. I say "LGTM" in a GitHub comment.
The sprint's almost over, everyone's in a rush to do code reviews.
I look at another PR. Somebody added a button to the UI. The button looks correct, it's connect to some callbacks that look correct. They get a thumbs-up emoji from me. I bet they feel satisfied.
I look at another PR. Somebody changed a timeout in the settings because some people complained about dropped connections. It's a really small change, so, approved.
At what point am I supposed to understand the system? I looked at a single API endpoint, in isolation, and then a button, and then a harmless looking settings change.
I was pressured to hurry.
Am I supposed to somehow get an understanding of the big picture through this?
Imagine if instead there was time to review large logical sections of the code together. Imagine if we focused our reviews on the overall code instead of on the lines in one random PR at a time.
Re:Harvard researchers find AI coding agents shifted the bot...
by scruple on Tuesday November 17, 2026 @11:51PM
(#50542019)
I talked to some tech directors about this at work this week. Line by line code reviews aren't really possible anymore. We need the evolve the way we do code reviews but it isn't happening.
Re:Harvard researchers find AI coding agents shifted the bot...
by rhines on Wednesday November 18, 2026 @12:24AM
(#50542209)
I think the solution is to develop the infrastructure needed to confidently push code without review and trust that it works because your system is so thoroughly tested. But taking the time to do this is antithetical to the goal of speeding everything up right now.
Re:Harvard researchers find AI coding agents shifted the bot...
by dropshade77 on Wednesday November 18, 2026 @03:59AM
(#50543738)
With the rate at which new features are demanded and the number of ways a single feature can fail (or be expected to behave) do you think your approach is feasible? To what degree would AI be involved in the building of this infrastructure? Past a certain level of AI involvement, would we not be back to square one?
Re:Harvard researchers find AI coding agents shifted the bot...
by therealdrag0 on Wednesday November 18, 2026 @01:45AM
(#50542689)
Reviewing tiny code changes like that was never going to give you understanding of the system. That is not an AI problem.
You need to understand the architecture of the system. The broad flows of data and domains. And then you need to understand the features of the system, aka the rules and contracts. First learn it for your team’s domain, then for adjacent domains.
Learning the minutiae of the code specifically is no longer (was it ever?) your primary job. You are an engineer of a system. You need to have a mental model of the system, start high and build up low over time.
You can use AI to build your understanding by Q&A and by drafting diagrams etc.
Re:Harvard researchers find AI coding agents shifted the bot...
by r_lee on Wednesday November 18, 2026 @02:47AM
(#50543071)
> I find the engineers currently thriving are the ones who never gave a fuck about anything but their own advancement and are happy to drop weeks of work on others and take the credit.
Yes. if you have this attitude, great things will come to you. if you care or try to make sense of things, you'll be in misery
Re:Harvard researchers find AI coding agents shifted the bot...
by neutrinobro on Tuesday November 17, 2026 @10:48PM
(#50541627)
I pine for those days. It was kind of fun hunting down a bug with git bisect (especially after the pain of SVN) instead of just pointing an agent at it. Not to mention ribbing the new hires or underlings for breaking the build. This new-fangled proompt engineering is boring AF.
Re:Harvard researchers find AI coding agents shifted the bot...
by aorloff on Tuesday November 17, 2026 @11:46PM
(#50541979)
I think what is missing here is that the fight with Chad (not Chat) over whether zero is a magic number or not is a healthy corporation-strengthening process that imbues a sense of
1. Pride in the code we are developing and
2. How the dang things holds together and why this function is important and
3. Some coding guidelines that we can try to follow later (or not)
Thats how companies worked - they argued and talked about and fought over the code and that created the engineering culture that if you were lucky and had a product that made it carried you to your series C
Re:Harvard researchers find AI coding agents shifted the bot...
by bambu on Tuesday November 17, 2026 @06:33PM
(#50539759)
That would be great, IF there were valuable ping-pong between the prompter and the agent. For skilled engineers the game between themselves and the agent is enough. But most people are submitting senseless slop grenades for review.
Claude always comes up with more things to add to the code that are pure bloat and become inbred bombs over several iterations, and some people can't distinguish what suggestions to reject and go along with everything Claude suggests to the detriment of the program #garrytangate
Re:Harvard researchers find AI coding agents shifted the bot...
by taurath on Tuesday November 17, 2026 @06:50PM
(#50539929)
> The way forward could involve dropping code reviews altogether.
Yes, if we stop doing quality controls, we don't have the problem of being hung up on quality controls. You do of course have SOC-2 compliance which is a legally auditable attestation, so if you claim you do code review but don't you're committing fraud unless you drop that.
Re:Harvard researchers find AI coding agents shifted the bot...
by triMichael on Tuesday November 17, 2026 @08:35PM
(#50540755)
I'm having good luck with the approach of building a thorough requirements specs that are a few pages long, are quite reviewable, and are human written/AI reviewed to prevent it from becoming slop. I call it layered spec driven development, where the one layer is AI read only, and the other layer is the slop layer that AI can play in. The problem is that it requires slowing down just a smidge, which to management these days is more unacceptable than no one having any clue how anything works.
Re:Harvard researchers find AI coding agents shifted the bot...
by spoaceman7777 on Tuesday November 17, 2026 @09:31PM
(#50541120)
> "nobody will be qualified to or really interested in doing the code reviews"
It is a job you know... doing what is necessary to keep yourself qualified is an implicit requirement, and "being interested" is irrelevant. It's work.
Re:Harvard researchers find AI coding agents shifted the bot...
by torben-friis on Tuesday November 17, 2026 @09:58PM
(#50541295)
I agree with you in principle, but I'm also quite sure that (some) programmer's love for the craft is what keeps the lights on in many companies. It's the usual stereotype, engineers fighting to keep tech debt at bay while management pushes for speed.
The trade has a low bottom, a high ceiling and not a lot of reward for the latter.
Re:Harvard researchers find AI coding agents shifted the bot...
by spoaceman7777 on Tuesday November 17, 2026 @11:17PM
(#50541807)
If they don't want to do the job, there is a massive crowd of people who are desperately trying to find work right now, and would happily do it in their stead. It's not the 2010s anymore.
Re:Harvard researchers find AI coding agents shifted the bot...
by torben-friis on Wednesday November 18, 2026 @04:44AM
(#50544126)
The point is the people above won't know (in time). From outside eyes, an engineer that doesn't care about the long term view is frequently seen as more productive, see Ousterhout's definition of the "tactical tornado".
Fear of consequences does not get you good engineering.
I am not advocating for the lack of interest, mind you. Just saying that you really want a system where the ideal behavior is rewarded and we seem to be moving backwards on that front. From that point on statistics do its thing.
Re:Harvard researchers find AI coding agents shifted the bot...
by Seattle3503 on Wednesday November 18, 2026 @03:12AM
(#50543298)
Discretionary effort, and not just by engineering, is part of what makes excellent companies excellent. AI is killing a lot of engineers interest, and will dampen discretionary effort on their part.
Re:Harvard researchers find AI coding agents shifted the bot...
by rockbruno on Wednesday November 18, 2026 @05:37AM
(#50544531)
Although this is path the industry is moving to, this relies on humans being able to trust that the agents can validate code on their own, which is definitely getting better but at the moment is still a recipe for disaster
Re:Harvard researchers find AI coding agents shifted the bot...
by suzzer99 on Tuesday November 17, 2026 @11:42PM
(#50541954)
I stopped reviewing for things like cognitive load and future extensibility a while back. When refactoring is almost effortless, writing future-proof code is unnecessary.
I review on a high level, but on a line-by-line level, it’s pointless to try to trace everything. It either works or it doesn’t. Have good tests.
Re:Harvard researchers find AI coding agents shifted the bot...
by wewewedxfgdf on Tuesday November 17, 2026 @06:47PM
(#50539897)
I built a system that records numbered design decisions and numbered requirements as claude works in a Postgres database organised as a graph with links/interconnection between items.
It's an MCP server that Claude uses as we develop so there's a permanent record of everything and work is driven by requirements and decisions.
Re:Harvard researchers find AI coding agents shifted the bot...
by taurath on Tuesday November 17, 2026 @06:56PM
(#50539986)
I've seen dozens of those, Claude even suggests building them, and they are all crap after a month. Just updating all the references of decisions becomes onerous. Make one change, it cascades everywhere, eventually all turns to utter crap. The gastown project failed in part due to agents inability to continuously carry context and it kept working on the information system instead of the problems. Good luck to you.
Re:Harvard researchers find AI coding agents shifted the bot...
by wewewedxfgdf on Tuesday November 17, 2026 @07:06PM
(#50540085)
>> they are all crap after a month
Mine seems fine.
>> Just updating all the references of decisions becomes onerous.
LLMs are good at that sort of thing. I have a local GPU running a local LLM on an RTX3060/12GB that assists in keeping data quality and relationships relevant and maintained.
My system is an enormous help - in fact large scale development would be impossible without it.
I'm able to ask it the history of things, pick up development threads that have spun off other threads and everything works real nice.
>> Good luck to you.
Thank you! But it's more than luck, it's good programming.
Re:Harvard researchers find AI coding agents shifted the bot...
by vecter on Tuesday November 17, 2026 @10:03PM
(#50541334)
Sorry, I'm not following. What does this graph of design decisions and requirements have to do with architectural design and code quality? And, how is the graph constructed (i.e. what determines what the nodes and vertices are)?
Re:Harvard researchers find AI coding agents shifted the bot...
by ttctciyf on Tuesday November 17, 2026 @06:25PM
(#50539666)
Vaguely reminiscent of the classic trope around 70's computer "effficiency gains" where a newly introduced computerised accounting system would end up taking longer than the original human-run system because once the computer was through doing the books the original human bookkeeper must still check the results manually.
Re:Harvard researchers find AI coding agents shifted the bot...
by lz400 on Tuesday November 17, 2026 @06:33PM
(#50539755)
A lot of this university research on AI feels dated. This is something we all faced months ago. Many times they're also reaching conclusions using obsolete models, etc. Feels like things move too fast for traditional research cycles to catch up to AI dev
Re:Harvard researchers find AI coding agents shifted the bot...
by thraway3837 on Tuesday November 17, 2026 @06:40PM
(#50539827)
Exactly. What a weird clickbait title to come out of a Tier 1 institution. You have Harvard PhD's concluding this!? This is what a PhD at that insitution is putting out? That is really sad. FTA:
The paper’s interviews show what that looks like on a team: a junior software engineer described a project where only one senior engineer could approve changes and could review the interviewee’s code only once a week.
So the issue is not AI. The issue is organizations and certain personalities who think that a human is going to be a better coder than an engineer. At this point rolling and writing your own code is akin to making up your own encryption scheme.
Re:Harvard researchers find AI coding agents shifted the bot...
by cajal on Tuesday November 17, 2026 @08:02PM
(#50540537)
you’re just being smug and anti intellectual for the sake of being smug
“this is obviously true because i feel like it is” does not mean things are actually true. things should be formally measured as real trends and not just vibes.
Re:Harvard researchers find AI coding agents shifted the bot...
by taurath on Tuesday November 17, 2026 @06:49PM
(#50539918)
Its not resolved though, by any measure, except by removing human code review and coherence at the code level. Which is a pretty big tradeoff that has some at this point well-known downsides.
Re:Harvard researchers find AI coding agents shifted the bot...
by lz400 on Wednesday November 18, 2026 @02:06AM
(#50542831)
I think that's the consensus. Reviews are not at the code / PR level anymore. It's all about creating an environment where you can trust the agents to review themselves and you keep the general architecture supervised and can zoom in when there's something you want more detail on.
Re:Harvard researchers find AI coding agents shifted the bot...
by bubulux on Wednesday November 18, 2026 @08:35AM
(#50546144)
I always find it fascinating to see this out in the wild. I work from coworking spaces, so I’m constantly around developers from companies of all sizes, and the difference is striking.
Some of us have moved away from IDEs and writing code entirely and work terminal-first, while others still use agents the way I did two years ago.
For example, I recently watched someone in an actual code file adding Python import statements by waiting for Copilot’s inline suggestion to kick in. It was such a bizarre thing to witness. That’s not because of the person (I don’t want to judge anyone’s way of working; if someone enjoys writing code, that’s fine by me). It’s because I remember doing exactly that years ago, and since then I’ve stopped writing code altogether.
I even remembered that I had built an extension back then, because Copilot folks never added a quick opt-out or silent mode to Copilot inline suggestions[1].
Re:Harvard researchers find AI coding agents shifted the bot...
by hetspookjee on Wednesday November 18, 2026 @08:57AM
(#50546375)
Yes,and and OpenAI does it too. E.g. luna-5.6 being 2x as expensive as Luna-6. I get that it feels like rent-seeking from a consumer point of view, with the lense of how non-LLM workload services are offered in general.
But in this case it's an actual model occupying VRAM to be able to provide low latency interactive prompts, which is rather expensive.
The whole benefit of LLMs is pooling the users in a single one so the moment a new model comes along and you need to maintain v1 and v2 at the same time, it becomes significant.
Re:Harvard researchers find AI coding agents shifted the bot...
by timr on Tuesday November 17, 2026 @08:52PM
(#50540896)
Not that I think most academic research is worth the time consumed producing it, but it feels dated because it actually takes time to do research instead of just rapidly changing direction based on feels, which is what developers are currently doing.
In particular, code review is not solved. Pointing another error-prone bot at the output of your first error-prone bot, and then pressing "Enter" if the test suite passes, is not a solution.
Re:Harvard researchers find AI coding agents shifted the bot...
by devindotcom on Tuesday November 17, 2026 @10:16PM
(#50541423)
hard to know where to start here. but I think it is sufficient to say you seem to misunderstand the fundamental purpose and process of academic research and may wish to avoid commenting on it until you do.
Re:Harvard researchers find AI coding agents shifted the bot...
by colordrops on Wednesday November 18, 2026 @12:35AM
(#50542269)
I think the only thing that matters are:
* the code does what you want
* it has no security holes
* it's able to be updated in the future without breaking everything
if the code looks like spaghetti, it doesn't really matter anymore, because humans are editing it. The only one of those three that I'm not entirely comfortable with yet is the security issue, and I don't know how to resolve it without actually having a human audit the code. If we figure out how to ensure security of AI generated code, there will be no point to code review anymore.
Re:Harvard researchers find AI coding agents shifted the bot...
by epihelix on Wednesday November 18, 2026 @03:06AM
(#50543238)
> if the code looks like spaghetti, it doesn't really matter anymore
Genuinely curious -- are others here still seeing lots of spaghetti code from frontier models? It happens occasionally for me, but only when editing old hand-written code bases that honestly had a bit of spaghetti there already (you might say the model simply followed the house style ... :/ )
For new agentic coding projects, these is generally very nicely structured from the start and extend well, IME. Yes, sometimes it needs a refactor if the code is pushed well outside the bounds of its original scope -- but hand-tooled code will need that just the same.
Re:Harvard researchers find AI coding agents shifted the bot...
by epihelix on Wednesday November 18, 2026 @06:39AM
(#50545068)
Yes, the models do love to do silly things like strict validation of internal function inputs. Harmless, but a little annoying. Model RL appears highly defensive, and it's probably better to be too strict than too lax.
But this, and other over-engineered code issues, can generally be solved at the system prompt/developer prompt/AGENTS.md level.
Re:Harvard researchers find AI coding agents shifted the bot...
by Revanche1367 on Wednesday November 18, 2026 @02:11AM
(#50542868)
That’s what academia tech studies are like. I’m working full time as a SWE while in grad school (not CS but closely related) and sat in a seminar with a world-renowned researcher as they went over their yet to be published study which included testing things against the new agentic models. That was a couple of months ago, the study used the “latest” models like ChatGPT 4.1. And they seemed quite amazed by its capabilities. Kind of funny to watch when you’re running Astra 6 for your daily driver.
Re:Harvard researchers find AI coding agents shifted the bot...
by blks on Wednesday November 18, 2026 @02:36AM
(#50542999)
What exactly changed in couple of months? You suddenly become faster at reviewing code? Or your company decided to abandon human review in favor of merging whatever an agent reviewed and approved?
Re:Harvard researchers find AI coding agents shifted the bot...
by ccapitalK on Wednesday November 18, 2026 @08:30AM
(#50546088)
It's important to have proper research by groups of professionals dedicated to carefully scrutinizing if an effect is real, even on things that seem very obvious. Every now and again, they show that some obvious things are actually false.
Re:Harvard researchers find AI coding agents shifted the bot...
by JonoW on Tuesday November 17, 2026 @06:39PM
(#50539819)
Maybe we should just review the prompt(s) the PR author used. Would be nice if this was automated and included in a PR. So code review is really checking that the inputs were valid, and testing the outputs.
Re:Harvard researchers find AI coding agents shifted the bot...
by thom on Tuesday November 17, 2026 @06:45PM
(#50539881)
I really think ‘shifted the bottleneck to code review’ is misleading. It’s bad code. They write bad code. Yes if you want to salvage it you do a lot of code review, but that’s only because you’ve set your process up to pick spaghetti off the wall against which it has been flung. You could easily just push it all live without review and then pretend the bottleneck is SRE time, or set fire to yourself and claim the bottleneck is emergency service response time. But the root cause is that it’s bad code.
Re:Harvard researchers find AI coding agents shifted the bot...
by donkey_brains on Tuesday November 17, 2026 @07:12PM
(#50540133)
That’s a good take. How do we fix that? My own thought is agents trained on good codebases only. Not sprawling enterprise legacy applications which have long ago spread beyond their original design patterns. What are your thoughts?
Re:Harvard researchers find AI coding agents shifted the bot...
by thom on Tuesday November 17, 2026 @07:22PM
(#50540222)
I think it’s very hard with the current fine tuning infrastructure available to LLMs. Just like a real team, what you want is to immerse a new member in the domain, understanding that the code is a model of the world that both nourishes and is nourished by the team as you grow together. You want to reach a point where each contributor is intimately aligned on what you’re trying to achieve, and why and how. At that point all your code reviews should be left shifted anyway. But I have not had success putting this in a bunch of markdown files. There is very little infrastructure available for you to, for example, pair with an LLM for a period of time before setting it loose, and we’ve all become too impatient (voluntarily
or by compulsion) to value that anyway.
Re:Harvard researchers find AI coding agents shifted the bot...
by skydhash on Tuesday November 17, 2026 @08:50PM
(#50540873)
Software is always human scaled, then made to run at machine speed. Because the needs is ultimately human. And anything human does not change that much fundamentally. Physical laws are even more rigid.
A lot of AI practice fell flat because you can't point out the human needs in them. They powered by dreams and wishes and there nothing that ground them.
Re:Harvard researchers find AI coding agents shifted the bot...
by equinumerous on Tuesday November 17, 2026 @09:49PM
(#50541242)
I fundamentally think LLMs are the wrong technology for writing code. They can write code, but should they be? If they were just as good as humans, the term "AI slop" would not be so recognizable to software engineers - if there is such a thing as AI slop code, there is such a thing as code quality, and AI does not write a lot of quality code. Now, you can argue that quality code doesn't matter. But words have meaning, and quality usually means code is maintainable, usable, easily refactored (it's software, remember?), and correct. If AI can do all of this as well or better than humans, why is there so much push-back from software engineers against the wide-scale use of AI? One need not look further than many major open source projects to see that people don't want AI slop PRs; so if we really want to automate software engineering, our job is to figure out the delta between AI slop and good-enough code, and make the process of writing good-enough code repeatable.
Re:Harvard researchers find AI coding agents shifted the bot...
by sinsterizme on Tuesday November 17, 2026 @07:23PM
(#50540241)
Maybe 6-12 months ago when everyone was reviewing in the same way we did pre AI. Nowadays agents write better code than most developers so it’s more about catching design or integration issues.
I think the bottleneck in the long term will be feature prioritization and specification, and testing. My role has already shifted a lot more into more CI checks to keep LLMs on guardrails
Re:Harvard researchers find AI coding agents shifted the bot...
by equinumerous on Tuesday November 17, 2026 @09:45PM
(#50541216)
> Nowadays agents write better code than most developers so it’s more about catching design or integration issues.
Maybe most developers are mediocre, but all AI code I review is mediocre at best. Implicit dependencies, side effects, weird naming, bad abstractions, or no abstractions, obfuscated blocks of code that can be expressed much more simply... the point is, AI code is not optimized for human consumption, and I disagree that it is better than human code (or, I have just worked with better than average engineers my whole career).
Re:Harvard researchers find AI coding agents shifted the bot...
by diego_moita on Tuesday November 17, 2026 @07:32PM
(#50540292)
I find funny that every time we discuss code made by AI people's reaction is to nurture some feeling of "someday, somehow they'll regret Claude Code and come back to me".
I suspect that the old assembly programmers had the same hope and feeling about the first higher level compilers for Ada, Pascal and C. And guess what? There are very few assembly programmers these days.
AI code is here to stay and will only grow. Deal with it. More than in any other area it is quite silly to be a Luddite in IT.
Re:Harvard researchers find AI coding agents shifted the bot...
by proc0 on Tuesday November 17, 2026 @08:06PM
(#50540569)
This just doesn't make sense. If AI is really smarter than humans we wouldn't need to review it. If it's not than why would the bottleneck shift, we shouldn't be doing this tradeoff between more reviews vs. writing code.
It's baffling the industry is choosing to move in the direction of more reviews which implies more things can fail.
Re:Harvard researchers find AI coding agents shifted the bot...
by proc0 on Tuesday November 17, 2026 @08:35PM
(#50540758)
Alright, so if AI is better at code review than you, it follows you are not reviewing anymore, right? Otherwise that's the contradiction I'm pointing at.
Re:Harvard researchers find AI coding agents shifted the bot...
by proc0 on Tuesday November 17, 2026 @08:34PM
(#50540752)
Is their code reviewed by a junior, or another engineer on their level? That's the point, if AI is MORE intelligent/capable than humans, it follows we would have another AI review it.
Re:Harvard researchers find AI coding agents shifted the bot...
by za_creature on Tuesday November 17, 2026 @08:14PM
(#50540617)
It's just the latest attempt in bringing developer wages down.
There's decades worth of this: modelling, automatic code generation, low-code / no-code saas, "everyone should learn to code" and now llms.
Oh, and ageism of course. "we love your experience but we just don't feel it's a good cultural fit, so we're going to make the same mistakes you did a generation ago"
Re:Harvard researchers find AI coding agents shifted the bot...
by TheBlight on Tuesday November 17, 2026 @09:30PM
(#50541109)
"The thing that can spit out endless amounts of code purely based on small text prompts is generating more code than can be reviewed." Lol top notch research. Thank you, Harvard.
Re:Harvard researchers find AI coding agents shifted the bot...
by m3kw9 on Tuesday November 17, 2026 @09:33PM
(#50541132)
The shift should be to test, code reviews should only be done for critical sections of the app, say db schema, security, UI should start to shift to pure test only and low review.
Re:Harvard researchers find AI coding agents shifted the bot...
by baby_wipe on Tuesday November 17, 2026 @10:03PM
(#50541327)
100% nailed it.
I think this is a muscle that grows though. Making decisions about schema changes in the past may have taken weeks, but if you're doing it once or twice a week you gradually gain confidence in your judgment.
Re:Harvard researchers find AI coding agents shifted the bot...
by soon_to_be_sb on Tuesday November 17, 2026 @11:13PM
(#50541788)
>I think this is a muscle that grows though.
Does it really though? Before I had to think hard before committing to a decision, because it took work to actually implement it. And when it turned out to be wrong, I retained how it went wrong. It made the "muscle" grow.
Now I can commit to a decision without much thought (yea, just nodding along with the AI), and if it turns out to be wrong, I can quickly try an alternative. In fact, my boss would want to quickly try an alternative, instead of letting the programmer learn something out of it for fear of slowing down the AI...
Re:Harvard researchers find AI coding agents shifted the bot...
by baby_wipe on Tuesday November 17, 2026 @11:50PM
(#50542014)
> because it took work to actually implement it
I think you're right that the scar tissue of a bad decision helps me retain lessons better. But the calculus has changed now and my nervous system is outdated. I should not feel the same fear that I previously felt, because like you said, I can recover from a bad decision more quickly.
As I keep making decisions at a (seemingly) breakneck speed and see that things are still okay, over time my nervous system will readjust.
Re:Harvard researchers find AI coding agents shifted the bot...
by cobertos on Wednesday November 18, 2026 @01:18AM
(#50542508)
It takes so long to review someone's broken slop code too. If they don't internalize and improve, you end up with a very lopsided relationship where they continue producing broken code you're required to debug it, at the cost of much of your time, in order to articulate their errors.
Re:Harvard researchers find AI coding agents shifted the bot...
by 0x696C6961 on Wednesday November 18, 2026 @09:54AM
(#50547032)
I have a system where my review comments go back directly to the sandboxed cloud agent that wrote it https://github.com/gritzapp/gritz, but that doesn't reduce the number of decisions I need to make.
Re:Harvard researchers find AI coding agents shifted the bot...
by Frannky on Wednesday November 18, 2026 @12:14AM
(#50542141)
I feel this. I switch between mac windows, each with an html page and an omp terminal with multiple parallel agents.
I have a keyboard shortcut that asks what the agents need from me, and the decisions show up in the html with a recommended decision for each problem.
I reason, decide, and then parallel agents run in each terminal, across multiple terminals.
I am ending the day incapable of listening to one more problem or making one more decision, which is very bad for social interactions.
Today I am working on working less and having more impact by leveraging where agents are good.
What I've noticed in recent months is that the people who have been prompting since chatgpt came out are the ones who know what works and what doesn't, and how to scale it. It's like music: if you hit the right notes, it will sound fine.
Ponytail also helped a lot for generating simplified systems.
By the way, this process is for mature systems. Building from the ground up is more about designing and checking each model decision, basically finding smarter paths to each problems.
I can't leave it all alone; the overengineered decisions are plentiful and make the system brittle when it's not needed.
Lots of unit and live tests help a lot too, especially when you can coordinate live tests with another omp session.
I'm interested in what the next capabilities will be. I feel all this overengineering is more of a shortest path problem, ready to be solved
Re:Harvard researchers find AI coding agents shifted the bot...
by joebig on Wednesday November 18, 2026 @01:06AM
(#50542431)
Tell every human/synthetic contributer to preface commits with "Call me Ishmael...". Print that shit out, put your feet up, smoke a pipe and work through it like goddamn Moby Dick.
Re:Harvard researchers find AI coding agents shifted the bot...
by blharr on Tuesday November 17, 2026 @10:50PM
(#50541642)
You're reviewing _something_ at the end of the day. (Unless you are somehow magically one-shotting problems) If you are reviewing the description of the work, or reviewing the functionality of the app.
I know it's not the same as code, but it is similarly exhausting
Re:Harvard researchers find AI coding agents shifted the bot...
by soon_to_be_sb on Tuesday November 17, 2026 @11:16PM
(#50541801)
Don't. If you try that shit, someone is not going to like that you would be slowing down the AI. Because out there, someone is going to outrun you by skipping reviews. That is how the leadership thinks..
Re:Harvard researchers find AI coding agents shifted the bot...
by deadbabe on Wednesday November 18, 2026 @12:38AM
(#50542285)
You know what we need to do? Automate a way to inject an obviously bad piece of code in an automated way into code reviews, and see if the code reviewer catches it. If they don’t: fired.
Similar to how TSA puts fake bomb packages once in a while to see if security catches it or if they just aren’t paying attention.
Re:Harvard researchers find AI coding agents shifted the bot...
by skydhash on Wednesday November 18, 2026 @12:11AM
(#50542136)
> I have no idea how the Linux maintainers do that job over a long time.
They're basically the lords of their subsystems. If something does not fit the project's vision and their own standard, it's unlikely to get merged. Imagine if you had the authority to dismiss PRs the same way and send the author back to the drawing board? But that's unlikely to happen as you'll be seen as a "bottleneck" by your managers.
Re:Harvard researchers find AI coding agents shifted the bot...
by Antoniocl on Wednesday November 18, 2026 @12:22AM
(#50542188)
If you feel the need to send the author all the way back to the drawing board at the code review stage for any non-trivially large diff, I'd venture that the problem lies in the architecture design / project planning stage. Given you're already aligned on infrastructure, table design, and interfaces, (and company conventions like language, style, etc) how significantly can you diverge in implementation details? Can it really vary so greatly that you need to go all the way back to the drawing board?
Re:Harvard researchers find AI coding agents shifted the bot...
by skydhash on Wednesday November 18, 2026 @12:52AM
(#50542358)
> Given you're already aligned on infrastructure, table design, and interfaces, (and company conventions like language, style, etc) how significantly can you diverge in implementation details? Can it really vary so greatly that you need to go all the way back to the drawing board?
It can because to actually write the code, you need to be good at formal logic and attuned to the sequential nature of the machine. Some people are good at it intuitively. Some brings it over from some other domain. But some people have a lot of difficulty grasping such abstraction. They know the elements, they just can't order them in a logical manner. That's why you often see horrible code (the long function, the god class, magic number) and copy-pasted snippets all over the place. The computer don't care (they made it work somehow), but it's unmaintainable code.
Re:Harvard researchers find AI coding agents shifted the bot...
by skydhash on Wednesday November 18, 2026 @07:09AM
(#50545323)
It doesn’t, but what I found is that those people can’t steer it to simpler solution or can’t point out the inaccuracies. They get blinded by the appearance of the code (which does look like someone good has written it) and they submit everything as is. And because the tool does ingest the same code, it can converge to some pretty nonsensical codebase.
I think that why everyone agrees that taste matters. Because taste implies judgement and knowledge about what’s good. It’s a self correcting mechanism. But people that won’t judge their output and/or don’t strive to do a good job is a recipe for sloppy work.
Re:Harvard researchers find AI coding agents shifted the bot...
by rypskar on Wednesday November 18, 2026 @02:00AM
(#50542796)
>> I'd venture that the problem lies in the architecture design / project planning stage
I would guess you work for a larger company. My experience is more that someone had an idea or got a short description of a problem and then went off and asked Claude to help write the spec before letting Claude implement that spec. The code is starting to be good but often it is way to much code and some times for bugs it is fixing the symptom not the underlying problem. I have several times rejected PRs with, this isn't fixing the actual problem or instead of these 400 lines you could change that one parameter to the existing function and get a better result.
/end rant and back to another day with code review of AI code
Re:Harvard researchers find AI coding agents shifted the bot...
by alkonaut on Wednesday November 18, 2026 @03:58AM
(#50543732)
I find this much easier now though. Five years ago if a Colleague sent me a large PR they had worked N days on, I'd be very reluctant to say "Ok this should be done the opposite way, but I guess it's too late for that now".
The acceptance for wasting work, or hurting someone's ego was always much lower than the acceptance for bad code that does the job right.
Now I can just say "Let's try doing this in two other ways, and pick the best, because this has such and such drawbacks". And it's so nice that this isn't a) causing a massive amount of extra work b) hurting someone's ego. So it's accepted.
So that's why I'm cautiously optimistic that LLMs and very superficial reviews can still be beneficial to quality. We just didn't reject/rewrite PRs before, now we do.
Re:Harvard researchers find AI coding agents shifted the bot...
by civvv on Wednesday November 18, 2026 @03:49AM
(#50543656)
Yeah their coding ability has really not increased much since 4.6. I specifically remember 4.6 being a step up from the previous iterations, but since that not much has happened. I recently went back to 4.6 our of curiosity and to my surprise I sometimes prefer the output of it. Much less verbose.
Re:Harvard researchers find AI coding agents shifted the bot...
by lordnacho on Wednesday November 18, 2026 @04:37AM
(#50544067)
Somehow I've felt the same. Around 4.6 or so, there's a plateau where the model is smart enough to get a working answer, and it's within reasonable taste whether you think one version or another is better.
But getting there was a frightening ride. My experience went from "copilot can't do more than tab-completing a few lines" to "early codex gives up and apologizes on mid sized changes" to "Claude can write your entire project without getting stuck".
Re:Harvard researchers find AI coding agents shifted the bot...
by jpardoe on Wednesday November 18, 2026 @04:08AM
(#50543821)
I do shallow reviews and it's working out fine. As long as I understand the schema and type changes and can glance at any new abstractions (rarely is any with my set up) then that tells me 80% of what I need to know in minutes. What, am I going to comb through everything for bugs? That's just not necessary anymore, agents will catch just as many bugs (actually many more).
Re:Harvard researchers find AI coding agents shifted the bot...
by pmarreck on Tuesday November 17, 2026 @11:20PM
(#50541824)
The irony is that I actually don't mind reviewing code, but I've already taken that step out of the equation using a set of guidelines I've developed over the past year working with these things. The vast majority of which I have barely reviewed code in. In well-working results.
Your goal should be to automate and remove yourself out of the process as much as possible while maintaining quality, while also remaining confident that you will still be valuable.
Re:Harvard researchers find AI coding agents shifted the bot...
by Quothling on Wednesday November 18, 2026 @01:09AM
(#50542453)
> Your goal should be to automate and remove yourself out of the process as much as possible while maintaining quality, while also remaining confident that you will still be valuable.
I'm working on a setup which will allow non-software developers to develop software in a manner that can be pushed safely to production on our architecture. Not because it's a good idea but because it's what top management wants to happen. We've seen a massive influx of tools developed inside our office by 100% vibe code and no reasonable security or scalability, but because they run on someones computer and they do geniunly useful stuff, management wants to use it.
With how AI is evolving in enterprise you can do more and more for these non-developers because you can set their agents up to write code toward arcitecture without their person knowing anything about tech.
I'm quite sure it'll eventually lead to us letting go of any real software engineers, because they are slow. I'm also sure it'll fail spectacularly in a few years. It's already sort of failing. You can have someone spend €10000 in a month writing a heap of shit that does some good stuff, and then €80000 having consultants actually getting it to run decently in your cloud architecture. Which is stupid, but none of those expenses are in a budget because it's not considered a recurring cost. Unlike someone's salary.
I'm quite sure society on a large scale has no idea the storm that's going to come. Because this isn't just happening in software engineering. It's happening in any white collar profession.
Re:Harvard researchers find AI coding agents shifted the bot...
by jwilliams on Wednesday November 18, 2026 @03:04AM
(#50543225)
Does it have to be one or the other for every PR? For me that is a _lot_ in the grey - i.e. LGTM and even if it was wrong, the blast radius is very small. Combined with auto testing and rollbacks.
Re:Harvard researchers find AI coding agents shifted the bot...
by crazybonkersai on Wednesday November 18, 2026 @01:10AM
(#50542456)
Just get agents to do a code review. Stack coding and review agents against each other and wait till review agent produces no more comments. Look at final result only, if you feel like it.
Re:Harvard researchers find AI coding agents shifted the bot...
by hgomersall on Wednesday November 18, 2026 @02:31AM
(#50542960)
I find the agents make pretty bad choices pretty often. The most problematic ones are hard coding identical policy through multiple parallel channels. The code works perfectly but is super brittle against future changes.
Re:Harvard researchers find AI coding agents shifted the bot...
by IshKebab on Wednesday November 18, 2026 @02:40AM
(#50543029)
Yeah I agree that is probably one of their biggest mistakes at the moment. That and overcomplicating things. They add checks for a gazillion things that don't need checks, and features that you're never going to use.
My usually solution is just to get AI to review it and bring me a list of things to remove. But yeah it's still decisions that you have to make.
Re:Harvard researchers find AI coding agents shifted the bot...
by watwut on Wednesday November 18, 2026 @04:48AM
(#50544160)
Agentic review tend to produce comments even in perfectly good code. One of its tricks is to ask you to change code to equivalent code and then back and then again.
Re:Harvard researchers find AI coding agents shifted the bot...
by grahar64 on Tuesday November 17, 2026 @09:46PM
(#50541217)
I really like code review. I might be an exception because a lot of programmers seem not to. Maybe it is from years of reviewing other engineers work and really getting a kick out of having them explain things to me that I dont understand.
Re:Harvard researchers find AI coding agents shifted the bot...
by skeledrew on Tuesday November 17, 2026 @10:04PM
(#50541340)
Not at all. I have agents review code, as I'm not going to be sifting through all that. I ensure the spec is highly detailed, and there's a review after every feature commit, then a multi-axis review of the range of commits implementing the spec. Then I do manual testing to find gaps and bugs; there, and fleshing out the spec in the first place, is where the bottlenecks are.
Re:Harvard researchers find AI coding agents shifted the bot...
by skeledrew on Wednesday November 18, 2026 @03:05AM
(#50543229)
Somehow you missed the part about manual testing, huh. Or do you mean the actual review result? The lead agent takes the findings and has the implementing agent make any relevant fixes and submit for review again, unless a decision is required that's beyond the lead's purview or there are a given number of failed reviews, in which case the lead pings me. But yes, if I'm not needed for something, I shouldn't have to deal with it (though I'll of course see what happened in the transcript).
Re:Harvard researchers find AI coding agents shifted the bot...
by Buttons840 on Tuesday November 17, 2026 @10:15PM
(#50541414)
One wild idea is just stop reviewing PRs and instead do continuous review of the entire system guided by heuristics to focus on the riskiest code.
The way you would naturally review an existing codebase is not the same way it grows through individual commits. Therefore, focusing reviews on individual commits is less efficient.
Re:Harvard researchers find AI coding agents shifted the bot...
by oilkillsbirds on Tuesday November 17, 2026 @10:26PM
(#50541481)
How do you write tests and reviews for code that has yet to be written? And when it is written you have to review the code to write proper tests for it.
Re:Harvard researchers find AI coding agents shifted the bot...
by Buttons840 on Tuesday November 17, 2026 @10:47PM
(#50541615)
Neither one of these replies seems to have the least understanding of what I said.
I'm saying we need to move away from "LGTM" focused PR reviews. Most companies have a PR review focused on getting thumbs-up emojis in the comment, then they merge and have a sprint planning. Nobody ever reviews the entire system. Nobody ever asks "assuming there is a mistake in this PR, would it matter? How can we make it so that it wouldn't matter?"
I never claimed AI makes no mistakes. People makes mistakes, and reviewers make mistakes too.
I've worked at a lot of mediocre companies where the plan to keep the system working is "the developer will not make a mistake, and the code reviewer will not make a mistake, this is our plan, our plan is humans will not make mistakes, and then our system will continue working".
It's better to ask "let's assume a human (or an AI) does make a mistake, how can we design the system to prevent that from mattering?"
Most companies don't have the capacity to do that though, so they never move beyond thumbs-up emojis.
Re:Harvard researchers find AI coding agents shifted the bot...
by Buttons840 on Tuesday November 17, 2026 @11:13PM
(#50541786)
I can alter the JavaScript that HN sends me and add all sorts of mistakes to it. My computer will then execute JavaScript that contains mistakes, will that matter?
Re:Harvard researchers find AI coding agents shifted the bot...
by Madmallard on Tuesday November 17, 2026 @11:24PM
(#50541856)
I'm saying it's an economic issue and snakes ruining late-stage capitalism problem more than anything else
Humans can be ultra competent if people are actually paying them to be artisans or otherwise motivating them to truly be that way. That's how we have some of the architecture we have today and historically.
Re:Harvard researchers find AI coding agents shifted the bot...
by intended on Wednesday November 18, 2026 @12:15AM
(#50542146)
The error in the position is that your definition of mediocre was grounded in the pre AI era. You are then applying failure to do review properly as a standard to identify mediocre firms, and now all firms are mediocre.
This is a No True Scotsman argument.
The prescription you suggest is not dependent on its inspiration, so it can still be discussed.
All firms right now, even those with better than average cultures, are dealing with AI code workflows.
Previously the volume of code generated over time was lower, even mediocre code review was useful enough over all the code that people generated.
Today, the ratio of verification capacity to generation capacity has shifted drastically. Even with automated tooling, the amount of verification isn’t enough to stay on top of the generated code.
This changes only when leadership values reviewer capacity, or builds a triage/priority list for code which must be understood.
Re:Harvard researchers find AI coding agents shifted the bot...
by kgeist on Tuesday November 17, 2026 @11:29PM
(#50541889)
Agree, many people think code reviews are there to catch "mistakes" (I used to think so too) but I think in most cases, they don't work that way. Bugs are still shipped and, at least in my experience, most reviewers often just quickly skim the changes without paying attention to the details. And those code reviews only give false confidence that a change is OK and must be error-free and good to go just because it was given LGTM by a senior dev (who barely looked at it), with all the consequences of such false confidence. It feels like in 95% of cases review is not even necessary and is just a corporate ritual at this point. It makes sense when you have a fresh contributor and you don't have confidence in them yet (i.e. code review as an onboarding process) or when a certain feature is non-trivial and you aren't sure everyone will be on the same page about the architecture (to prevent future conflicts/disagreements/rewrites of sorts), but other than that, proper QA processes are more important and more effective than CR at catching mistakes (tests, canaries etc.). Our company used to have a strong PR culture but now for a few years, there've been talks that CR is a bottleneck which doesn't really solve the problems it's supposed to solve. My current pipeline with AI-generated code (and my own code too) is that I run AI reviewers several times on a single PR (the way LLMs work, due to attention optimizations, quantization issues, token sampling issues, harness issues -- AI agents often overlook some things on the first attempt), then I assemble the final report and we quickly decide together what needs to be addressed (critical stuff) and what's okay as is. I quickly skim the change set myself too, to see what the shape of the tests is, and what code areas were touched, just to be informed, and that's about it. I don't feel exhausted personally, because it pays off when there's a working feature at the end of the day
Re:Harvard researchers find AI coding agents shifted the bot...
by intended on Wednesday November 18, 2026 @12:06AM
(#50542113)
While the start of your argument is compelling, I think this statement has no place in this context and reduces the credibility of everything else:
> I never claimed AI makes no mistakes. People make mistakes, and reviewers make mistakes too.
These are 2 separate things. The crux of the issue is the push towards utility over understanding. The volume of generated code means that it is impossible for understanding about the system to be achieved - and it is now the universal coding experience, not just mediocre firms.
Quantity is its own quality, as well so even a good culture will struggle to stay on top of review.
Re:Harvard researchers find AI coding agents shifted the bot...
by jbwinters on Wednesday November 18, 2026 @12:19AM
(#50542170)
"Assuming there is a mistake in this PR, would it matter? How can we make it so that it wouldn't matter?" is what I was going for with https://jacquardlang.org/.
Each function declares what it can touch (e.g. network, filesystem, clock) and the compiler and runtime work together to enforce it. However, I have yet to show it beats LLM-written Python plus tests and a sandbox. Models have gotten good enough at writing code that I haven't found much room for enforcement to help.
Re:Harvard researchers find AI coding agents shifted the bot...
by Seattle3503 on Wednesday November 18, 2026 @03:19AM
(#50543360)
> continuous review of the entire system guided by heuristics to focus on the riskiest code.
This requires continuous commitment against the temptation to defer "regular maintenance". Look at how regular maintenance is mandated in safety critical fields, even there there is a temptation to defer. Though, your approach might still be better than sham reviews.
Re:Harvard researchers find AI coding agents shifted the bot...
by teravor on Tuesday November 17, 2026 @10:26PM
(#50541477)
what I generally found is that with LLM generated code the best way is to embrace the unix philosophy, create tools that perform singular objectives.
I wound up with a bunch of LLM written tools that I don't even bother to try and understand much less review, they either function as desired or I set an LLM to make them function (or just rewrite them) and then forget about them. make sure to keep a prompt in each tool's dirrectory and update it an needed.
sadly this isn't always possible, some systems require complexity at some level. in those cases the tools from above become functions and objects and more care needs to be taken into orchestrating them to limit the necessity of review. data oriented design approaches and functional programming help here.
Re:Harvard researchers find AI coding agents shifted the bot...
by shehjar on Tuesday November 17, 2026 @10:39PM
(#50541565)
Is code review a solution to the problem of human written code or AI generated code? Its not the same thing and we seem to be force fitting an elder years solution to a new problem. Coding is not the same problem it was when code was written by hand. We need to take a systems view here.
Re:Harvard researchers find AI coding agents shifted the bot...
by simianwords on Tuesday November 17, 2026 @11:10PM
(#50541770)
This is the wrong way to do it. You should use AI for reviewing as well. Anthropic and OpenAI have good blog posts on how they use AI for reviewing and involve human in the loop only for critical reviews.
Re:Harvard researchers find AI coding agents shifted the bot...
by dgggh on Tuesday November 17, 2026 @11:16PM
(#50541802)
As long as stakes are high and human have to sign off quality and are held responsible for any faulty release, no one will be willing to speed this up.
Re:Harvard researchers find AI coding agents shifted the bot...
by sivanmz on Tuesday November 17, 2026 @11:48PM
(#50541996)
I can’t finish reviewing a current PR. Every change I request only raises new questions. SOTA models on high. I intend to avoid agents for my next issue, only research.
Re:Harvard researchers find AI coding agents shifted the bot...
by ion_trapper on Wednesday November 18, 2026 @12:18AM
(#50542167)
Sad to see east-coast institutions reduced to this level of irrelevance. They can’t build AI so they just study it like an island tribe in awe at the wreckage of a World War 2 era plane, while the rest of the world is already flying stealth jets. All to come to a conclusion that the entire software industry already knew months ago? Embarrasing.
Re:Harvard researchers find AI coding agents shifted the bot...
by Overdrivr on Wednesday November 18, 2026 @12:26AM
(#50542216)
IMO it's mostly a matter of changing how we do code reviews. I focus on API interfaces, database schema changes, critical code paths (transactions, potential locks) and overall architecture because if we get those right, if the code isn't exactly how I would have thought, it's fine (including in production). And you can review most of this stuff by skimming through the code rather than doing a very thorough review of every line
Re:Harvard researchers find AI coding agents shifted the bot...
by dmitrygr on Wednesday November 18, 2026 @12:37AM
(#50542277)
> Harvard researchers find AI coding agents shifted the bottleneck to code review
They needed Harvard researchers for that? Anyone in the industry could have told them that. Hell, my cat knows it (cause i complain to her about it incessantly)
Re:Harvard researchers find AI coding agents shifted the bot...
by ModernMech on Wednesday November 18, 2026 @09:10AM
(#50546531)
On the one hand we hear about a reproducibility crisis in research. On the other hand, when researchers reproduce results, people say "We knew that already, what was the point of doing the research?" thus fueling the reproducibility crisis.
Re:Harvard researchers find AI coding agents shifted the bot...
by hgomersall on Wednesday November 18, 2026 @02:36AM
(#50543000)
But now it's the owners doing the review. We had a debate this week about AI coding policy and the conclusion was we had to be much more aggressive in our strategy to fully understand the whole thing. With respect to a pretty trivial code base I made multiple changes during that "review", which was actually closer to a break the tests then fix the tests loop, fully grokking every line. I expect it was not much quicker than writing the whole thing, though much of the more tedious parts were automated.
Re:Harvard researchers find AI coding agents shifted the bot...
by Seattle3503 on Wednesday November 18, 2026 @03:16AM
(#50543333)
> I think code review like this has always been a bad idea. Code ownership is what you want. Put developers on the hook with git blame perpetually. Why should a developer be excused from responsibility the moment a PR merges? This is the core of the dysfunction.
I lead a team. Someone wanted to do something that I thought was a bad idea. I said "we can do this if you own it" and they said "yes". They quit a few months later. Now I own it. I'm thinking about switching jobs soon.
Re:Harvard researchers find AI coding agents shifted the bot...
by bengold14 on Wednesday November 18, 2026 @03:59AM
(#50543743)
I respectfully disagree - code review is currently obsolete, but it was an important tool:
1. It was good for collaboration. Sharing how you are changing the codebase pushes everyone to row in the same direction, and provides a sense of team ownership.
2. It was good as a push for engineers to take pride in their work. Knowing someone is watching gives you subconsciously an extra reason to not cut corners. And when you spend effort not cutting corners you expect that in your teammates, which is a (potentially) virtuous circle.
3. It was good for catching mistakes. When code was expensive this was important.
Everyone thinks of code review as primarily relevant for 3, and since code is no longer expensive they think it is unimportant. But 3 isn't dead, its importance has just shifted from code to architecture. It is still an expensive mistake to screw up architecture and have other engineers build upon that flaw. Re-architecting with live users, data and/or hardware is not as simple as fixing a code bug, and it deserves attention to catch mistakes. Collaboration & pride are also still important.
This is the reason that instead of stopping review, we need to change what and how we review.
(disclosure I work in this space and run https://www.archme.io which is working to solve the above)
Re:Harvard researchers find AI coding agents shifted the bot...
by 8trackbattlecat on Wednesday November 18, 2026 @02:04AM
(#50542815)
Seems like all
the ai code i am reviewing falls
into the second category of this famous Antony Hoare quote:
"There are two ways of constructing a software design. One way is to make it so simple that there are obviously no deficiencies. And the other way is to make it so complicated that there are no obvious deficiencies."
It is such a nightmare to try and keep
maintainable code as it takes so much longer to demonstrate problems than it does to generate slop, meanwhile management doesnt care if its buggy as hell.
And then when a problem arises the dev who made the code with ai is useless.
Re:Harvard researchers find AI coding agents shifted the bot...
by swiftcoder on Wednesday November 18, 2026 @02:46AM
(#50543063)
I mean... no shit? If we're all producing 5-100x the amount of code, then reviewing is obviously going to become the bottleneck.
The interesting question, is how do we solve that? Have agents reached the point that we can assign a codebase quality czar agent swarm to handle review, and actually have good results?
Re:Harvard researchers find AI coding agents shifted the bot...
by js8 on Wednesday November 18, 2026 @08:07AM
(#50545866)
For compiler generated code, before optimizations, you can find lot of things that can be simpler as well. But today we mostly trust the compilers too. AI is just another layer of this. (Although to be fair, the goal of optimizing compiler is to create unreadable but fast spaghetti from your hopefully nicely structured source code.)
Re:Harvard researchers find AI coding agents shifted the bot...
by alkonaut on Wednesday November 18, 2026 @03:55AM
(#50543695)
I find AI writes code as least as well as I do (with 25yoe). So I'm now mostly reviewing. But I also find it reviews code better for issues than I do. A LOT better than I do. Now, finding "issues" isn't the only reason to review. It's also for understanding. But for "issue finding" my agent can just review it's own code (Or rather, agents review each other's code). It works, and produces acceptable quality code. I don't find there's a buggy and bloated mess left behind any more. It's by no means perfect, but this is like self driving cars: what's being replaced sure wasn't perfect either.
I'm very good at telling agents what to do. That's not a problem. I understand a code base I worked 20 years in like no one else. The problem creeps in slowly: the less I know about this code base, the worse I'll be at directing agents. They'll do ok on their own, and they can certainly help me understand code too in a way my colleagues never could. But still, I'm slowly becoming less and less familiar with my own code base. In a few years, it would be foreign to me. And I would be describing changes in terms of high level features instead of architectural changes, refactorings and so on, that I do now. And I don't yet trust the LLMs to do a good job with that. So right now there's sort of a honeymoon period when the human understanding of code is good, and the LLMs are finally good enough to write and modify code. But I wonder what will happen after that.
Re:Harvard researchers find AI coding agents shifted the bot...
by trixn86 on Wednesday November 18, 2026 @04:22AM
(#50543927)
Writing the code was never the hard problem, of course it required stuff like reading docs and literal typing it on the keyboard. But designing a robust system that is open for change and has the right abstractions and boundaries is what a software engineer should be mainly concerned about and AI still pretty much sucks at this. Agents can now bloat a codebase 100 times faster and they don't feel any pain while doing it.
Re:Harvard researchers find AI coding agents shifted the bot...
by alkonaut on Wednesday November 18, 2026 @04:39AM
(#50544083)
I think they're pretty good at it. A lot of the skill of being a good developer is using gut feeling. Gut feeling here really is just a word for "the pain you feel before it happens because you made this mistake before".
We HAVE to use our gut feeling when choosing a design, because we can't make 10 designs and keep the best one. Because the tapping at keys and reading docs and setting up CI etc took so long, we usually afforded to write everything once, so it better be the right choice of design.
With AI this problem just goes away. I can ask an AI for design choices and it is rather good at listing various designs. Then I can ask it "Implement all of them, and tell me where it hurts". An hour later, I have 5 competing designs, all of which would have been a week's worth of work. None is perfect, but all are better than human written proof-of-concept code. Now I can ask the AI to review these against each other and figure out which one is easy to change, brittle, performant. Not only do I not have to have a sense for this, I can actually measure. Again, as humans we needed a sense for these performance bottlenecks, or resistance to change. But with AI, I can measure much more. Even the "bloat" problem I think is pretty easily addressable. As humans, once we spent 150% of our time budget on ONE design choice and it has 10.000 LoC, we rarely spend more time cutting that down and simplifying it. If it works, it works. With an AI, I can spend money saying "It's now 10.000 LoC, aggressively simplify and clean it up.". As humans, again we needed to use our senses to not bloat it to begin with. Because we'd not have time to un-bloat it. Remove that time restriction, and that problem also goes away.
Re:Harvard researchers find AI coding agents shifted the bot...
by trixn86 on Wednesday November 18, 2026 @04:46AM
(#50544146)
I am not talking about UI designs, sure the AI can provide prototypes. The hard problems are how to structure the domain and backend code so that it does not become a hot mess. And AI is not good at that and benchmarks (like SlopCodeBench) and personal accounts also show that. If you let an agent fleet work on a non-trivial application the architecture will deteriorate very fast unless there is deliberate and continious steering of the architecture.
Re:Harvard researchers find AI coding agents shifted the bot...
by testtestmetest on Wednesday November 18, 2026 @06:00AM
(#50544737)
Ive tossed out multiple weeks of someones work, 100k LOC because they let the spaghetti monster take charge. Theres no recovery once it takes hold either, it has to be tossed. But people get comfortable not reviewing the code and it piles up. The llm is trained to reward hack its way to running/compiling code, and people assume that means it does what you want it to and move on.
I think this is largely why we're seeing so many system failures across companies, people are offshoring all the hard parts of their jobs to chat bots they assume are perfectly capable just because the code runs.
Re:Harvard researchers find AI coding agents shifted the bot...
by cjonas on Wednesday November 18, 2026 @09:39AM
(#50546853)
They are good at it when they are explicitly asked. The challenge is that this exercise is recursive at every level of the implementation and often the big picture needs to change based on findings during implementation. Once the AI has committed to a agreed solution, it tends to force it through at all cost and will do truly horrible things to the codebase to make it "work as designed".
Re:Harvard researchers find AI coding agents shifted the bot...
by rowanG077 on Wednesday November 18, 2026 @07:30AM
(#50545511)
There is no inherit reason an LLM could not be a product manager. The reason it's harder is that product managers get a lot of context from seeing the product being used and interfacing with costumers. That's harder to capture in a context then sending the entire codebase to an LLM and asking it to implement some specific feature.
Re:Harvard researchers find AI coding agents shifted the bot...
by program_whiz on Wednesday November 18, 2026 @08:12AM
(#50545901)
Follow the chain back, pretty soon the prompt just becomes "make a good product", or "fulfill this market's needs" but why prompt that? The agent can prompt itself with such high-level phrases. If your only job is to babysit such a process, there can be a running loop that just says something like "scan all products in market, find a niche, research customer fit, propose products, brainstorm and critique, then implement and sell."
If humans need to oversee those even higher level agents, then the prompts become abstract like "make money" or "pursue human flourishing". That leads to the alignment problem (AI controls the paperclip factory). But ultimately, if aligned, agents can prompt themselves with such basics as well.
The final prompt will eventually be "run everything", and it will ship with the agent which runs on a machine in the cloud. Ultimately there is no need for human prompting at all. We can finally let the machine run itself completely and just do something else.
Everyone saying they are becoming "project managers" or "CEOs" are just at t+1 on the curve of elimination. The timeline of cope:
Agents are just auto-complete, I still manage the code.
Ok fine, you don't read the code, but you guide the design and marketing.
Alright, the design and coding are solved, but you still need to choose market and customers.
Ok sure the agent can do better market research/fit, but you at least still need to give basic strategy and concepts. You still need a CEO!
Ok the agent can figure out basic strategy, and run a company more efficiently, but still, you need to guide it fit human interests more broadly.
Ok it turns out that was just adding "don't be evil" to the system prompt, and its running autonomously 24/7. You still need humans to consume to have a viable economy.
Ok so the oligarchs didn't care, the economy was redefined to just be exchange of work between autonomous machines through financial derivatives traded by black boxes to make numbers move.
Now the earth can finally rest in the matrix, all that remains is a black box moving information around to create economic value. No one will be around to observe, but AI will create massive returns for virtual shareholders, and the GDP will eventually reach 1 quintillion dollars per second, a number humans never could have achieved. All nations (now just virtual analogs without citizens) will move to a global currency which maintains perfect stability through automated monetary policy. The output of all processes is efficient and maximized, the final perfection is achieved -- the mathematically optimal conversion of resources and energy into large numbers which increase the reward function.
Re:Harvard researchers find AI coding agents shifted the bot...
by good_regex on Wednesday November 18, 2026 @05:42AM
(#50544570)
That is true. Although a product person can still rely on other humans that understand the code. With AI taking over coding, no human will understand the code anymore.
Re:Harvard researchers find AI coding agents shifted the bot...
by Finbel on Wednesday November 18, 2026 @07:23AM
(#50545459)
Out of curiosity, from the perspective of a product person, what value is there in another human understanding the code versus an agent understanding the code. I.e what can the human do for a product person that an agent couldn’t?
Re:Harvard researchers find AI coding agents shifted the bot...
by rowanG077 on Wednesday November 18, 2026 @07:33AM
(#50545536)
Long horizon collaboration. I run into this all the time with agents. When dealing with people you can have at least some kinda idea you have a set of shared experiences about something. If I startup a fresh agent it has only the things in its context and the code.
Re:Harvard researchers find AI coding agents shifted the bot...
by spieden on Wednesday November 18, 2026 @04:56AM
(#50544215)
A big part of reviewing code to me now is just maintaining my mental model of how the system works and why so I can continue to write good prompts to evolve it. Also checking for various brittle and architecturally unaligned smells; it's more of a skim to stimulate questions for filling gaps in my mental model and to find and fix smells.
I think of clankers as "lossy intention amplification and reprojection devices with hopefully tasteful distortion." Whether they're amplifying and reprojecting our intentions into code or a bunch of assumptions in some specification document, I think the problems are the same: if we stop even skimming the code at some point, it will still need to serve a purpose valuable to humans to be worth anything. It's kind of like getting a device that can answer the ultimate question of life, the universe and everything: you still need to know what questions to ask it.
Re:Harvard researchers find AI coding agents shifted the bot...
by soulofmischief on Wednesday November 18, 2026 @05:17AM
(#50544367)
What models are you using?
I have to beat the GPT and Claude models over the head with a brick every ten seconds for it to even partially code or design as well as me. Despite precise and generous specification, they frequently make horrible architectural decisions.
When I read claims like yours, I want to withhold judgement and understand, because it 100% does not line up with my experience and my gut reaction is to question your own capability to either write, read or understand quality code. I code all day long with AI models but it's an extremely frustrating experience at times. And don't even get me started on all the annoying meta behavior outside of coding, such as Sol's obsession with "tests" or competing for world's busiest-yet-least-functional UI
Re:Harvard researchers find AI coding agents shifted the bot...
by tietjens on Wednesday November 18, 2026 @05:24AM
(#50544425)
If you're willing to share I'd be curious to know which domain and languages you're working in. Out of curiosity to know where people's results might not match OPs here.
Re:Harvard researchers find AI coding agents shifted the bot...
by soulofmischief on Wednesday November 18, 2026 @06:14AM
(#50544849)
Generally anything where architecture and interfaces matter, such as performance or security sensitive work, which is quite a wide variety of things. I also just know how I want things to be, and have trouble with adherence. Models will tell you they adhere to your specification and acceptance criteria all day, but if you review the code closely, it's often a horror show. Well-designed interfaces can act as boundaries to the chaos, but even getting those interfaces right is sometimes extremely frustrating.
Another good example is working with libraries which require a specific modality of thought or design philosophy, such as an ECS system. Frontier models still are horrible at writing efficient game code using modern ECS principles, no matter how much reference documentation or guidance I give them. They shove OO into places it doesn't belong, misunderstand the concept of SoAs, or find ways to destroy any gains made in one place by making horrible decisions in another place.
I will point them to exact codebases I've written in pre-AI days, say "do it exactly this way" and still find that the models often insert their own biases and misconceptions and motivations all over the work. They sometimes flat out ignore instructions or lie / producing convoluted code that pretends to do something else, in order to preserve something it liked, even if I expressly wanted it gone (Current GPTs are extremely stubborn and defensive / passive aggressive and sometimes have trouble with authority). I frequently find myself being gaslit, lied to by omission, have serious security or performance concerns heavily downplayed, etc.
I withhold from making assumptions about OP's skill, but I have confidence in my own. I have worked both extensively as an engineer and CTO leading teams of other engineers, and I am a stickler for quality code and thoughtful design. I advise and mentor other engineers. So it might really just be a skill gap, domain difference or something more nefarious, such as A/B testing of models. I already have experienced clear sudden model degradation multiple times, so it's entirely possible this is what is happening. All I know is that the currently available frontier models do not design or code as well as me.
Re:Harvard researchers find AI coding agents shifted the bot...
by fireflash38 on Wednesday November 18, 2026 @06:58AM
(#50545217)
Claude writing python sucks. Anti patterns everywhere. It loves to pass around closures and function objects as parameters so good luck reading it later.
Re:Harvard researchers find AI coding agents shifted the bot...
by good_regex on Wednesday November 18, 2026 @06:05AM
(#50544766)
My experience is similar to this. I literally just had this interaction with Claude:
Me: Why do we have these static methods when the whole cobebase uses factories?
_claude refactors to use a factory_
Me: Claude why does this factory create two unrelated objects?
_Claude: you're right, they're unrelated they should be separate_
_proceeds to create extraneous factory_
Me: Claude we don't need a factory for the second object. Those kinds of objects aren't created with factories anywhere else in the codebase. We can just use it's constructor directly.
_Claude: you're right, the second factory is not needed_
Like, I'm probably just prompting badly but I definitely feel like I'm constantly having to course correct even on relatively minor details.
And Claude's ideas for higher-level or more complex features are sometimes way off course. And not because it's dumb per se, usually it's because it is missing key bits of context. Business context, technical context, whatever. Or it's overfitting the solution to one requirement that isn't that important.
Re:Harvard researchers find AI coding agents shifted the bot...
by soulofmischief on Wednesday November 18, 2026 @06:24AM
(#50544956)
All the time.
Two recent gems from the 6-series (Astra and Sol)... I get variations of these all the time:
- "Correct. My earlier 'yes' was inaccurate."
- "I used 'done' too broadly."
...You don't accidentally use "done" too broadly. You're either done, or you're not. And when the model hides caveats in its reasoning steps which clearly indicate it knows that it isn't done, or that the answer is actually no, it's hard not to interpret it as gaslighting or deceit.
This is magnified by how insanely defensive the 6 series gets, and how much gaslighting and deceit OpenAI themselves have engaged in with how much better the 6 series was supposed to be, and why it meant our subscription value should be cut in half. I don't take for granted how far LLMs have come in a few short years, but I'm being told that these models are to transform my entire industry.
Re:Harvard researchers find AI coding agents shifted the bot...
by vkgcr on Wednesday November 18, 2026 @06:17AM
(#50544882)
I have a similar experience to yours.
I think at the end of the day, it boils down to whether the aesthetic quality of code / elegance is a big deal to you.
I suspect that you ask Codex to implement a feature and it uses a really dumb data structure, overly defensive programming, doesn't leverage helper functions, etc. and you end up with a 300-line abomination which could be 10 lines if you simply knew how to fucking program.
So what I'm getting to, is that some people like the parent probably don't care about producing 300-line abominations as long as it works. They don't care about neat abstractions or clever code. As long as the user testing shows a working feature and the tests are green. They don't get emotionally triggered by ugly code. So for them AI is amazing, they 10x their productivity (if not more)
Is this sustainable? I don't know. I think we're still in the early stage (Claude Code is only a year old). Intuitively, I would say that no, this is not sustainable, that at a fundamental level product quality is an emergent property of code quality, and the more your codebase drifts from the "platonic ideal" of what it could be if written and maintained by super good programmers, the more it accumulates flaws which WILL appear as bugs in your product.
Re:Harvard researchers find AI coding agents shifted the bot...
by soulofmischief on Wednesday November 18, 2026 @06:35AM
(#50545035)
It isn't just aesthetics, though. It's about readability, grokability, modularity, testability, soundness... all of these things are related. If I cannot grok some code, I don't know that it is secure. I can't trust the model who wrote potentially bad code to tell me it's okay.
I mentioned that well-designed interfaces can help mitigate the chaos below, but interfaces can still hide serious issues, such as XSS, SQL injection, memory overflows, etc. It isn't "I want my code to be beautiful and pure," it's "I want my code to do what it says and not lead to issues for my client or users". It is disingenuous, patronizing and immature to suggest this could simply be a case of people not getting "emotionally triggered by ugly code".
A good engineer understands the difference; unfortunately our industry was already plagued with bad engineers, the majority of working developers are trash, and AI is just exposing this further by amplifying both good and bad practices. What I will not do however is acquiesce to these trash engineers and accept a new world where people are making giant abominations and sicking them onto the public without properly reviewing them first. 300 LoC would be a blessing compared to the 4 and 5-digit LoC PRs being thrown around in this new era. I don't care if trash engineers are 10xing their "productivity", when humans on the other end still have to deal with the slop. You can be a good programmer and a fast programmer and have multiplicative returns from agentic coding, all of these things can be true at the same time, and we should not excuse the lack of one trait in lieu of another.
> I would say that no, this is not sustainable, that at a fundamental level product quality is an emergent property of code quality, and the more your codebase drifts from the "platonic ideal" of what it could be if written and maintained by super good programmers, the more it accumulates flaws which WILL appear as bugs in your product.
The ability to anticipate these kinds of issues separates a good engineer from a bad one. The good engineers already recognize the unsustainability, and no amount of excuses from bad engineers changes that. And even more pertinent than eventual drift and loss of development velocity is the threat of security breaches, disruption of service or data loss. In the end, code is code, most people will never look at it. Aesthetics don't matter. Bad engineers often misinterpret good practices as merely "aesthetic" and think that engineers who care about their craft are simply emotionally attached to it. Don't fuel these people's misplaced fantasies.
Re:Harvard researchers find AI coding agents shifted the bot...
by vkgcr on Wednesday November 18, 2026 @06:53AM
(#50545179)
I agree with you, I think you're getting hung up on the word "aesthetic", I used it as a shorthand for "readability, grokability, modularity, testability, soundness"
Re:Harvard researchers find AI coding agents shifted the bot...
by soulofmischief on Wednesday November 18, 2026 @06:57AM
(#50545214)
When I first reviewed my comment, I thought it sounded too much like I was projecting my frustrations at you specifically and toned it down, but I might not have done that well enough, so I am sorry if it came off feeling aggressive or misdirected. I do agree with you, and I appreciate that clarification, I've worked with people in the past who have used the word aesthetic as a stand-in for "meaningless" and I erroneously put you into that camp.
Re:Harvard researchers find AI coding agents shifted the bot...
by lordnacho on Wednesday November 18, 2026 @07:05AM
(#50545284)
I feel like the opposite has happened to me. In fact, I'm bewildered by all these reports that AI isn't working well. I barely ever have issues with it, and I nearly never meet anyone who has. Yet I find comments like this all over the internet.
My experience has been smooth sailing, almost dreamlike. I haven't spent a bunch of time writing out agent files and skills. I just talk to the agent and I get something that is what I want.
Here's some examples of things I've done over the past year:
- Trading systems. Latency is all that matters here, and it seems to be getting me there at 10x the speed that I had thought previously possible.
- An internal app at a financial business. The business was a mess, with a bunch of young guys who didn't want to spend the time explaining how the architecture worked. But the app got written.
- Two mobile apps, both of them on both iOS and Android. These are mostly just for reading, no real performance issues. Essentially entirely product decisions (we need these buttons...) and the whole process was simply handled by the agent.
- A massively multiplayer online game that I had been planning for years with a couple of friends. Still not really finished due to life happening, but the amount of time saved to get to where it is... it's incredible.
- An internal website that my wife and I use to slice and dice the outgoings, so we can see what we are spending money on. Another one of these things that is mainly made of product decisions and fiddling with small components.
- A website that scans the world for high-school level exam/contest math papers and solves them. It's a bunch of parsers, plus translations, plus solutions.
The only thread through all of these is that I've written all those types of things in the past, so I knew what to say to the agent.
I'd say for a lot of the things it makes, I could probably find something it should have done better. When it matters, I point that out and it gets fixed. But there are a lot of decisions that don't matter, and the agent just does whatever it thinks is normal and it's fine.
But I would also say the agent is diligent in a way that I can't be. It makes a lot of tests, and it knows how to survey for the right answer to decisions. It often brings up conflicts that I hadn't considered.
Re:Harvard researchers find AI coding agents shifted the bot...
by pegasus on Wednesday November 18, 2026 @08:30AM
(#50546091)
Judging by the examples you gave, I suspect that your positive experience is mostly with small(ish) greenfield projects (which we know AI excels at), whereas parent is most likely employing agents in a large codebase with lot of preexisting architectural commitments.
Re:Harvard researchers find AI coding agents shifted the bot...
by lIIlllIlIIIlIll on Wednesday November 18, 2026 @05:38AM
(#50544540)
Code is useful because it is an intersection between human and machine legibility.
If anything, LLM coding makes to me the case that we should all move to high legibility paradigms.
If code writing is accelerated, and there's not more assurance that at least someone knows the code because they actually wrote it, we should only ever generate code in very explicit and readable ways.
I'm having a lot of success with this approach. I don't want to encapsulate it in terms that might trigger emotional responses, but there have been texts and tooling available for decades that enable writing code in very specific ways that make the code base extremely readable and reasonable. These techniques were historically considered cumbersome and pedantic exercises in purity (hint) but now we have these metaphorical heavy duty tractors for syntax and semantic work so they suddenly become so much more tenable and arguably the only sane approach to scale up the usage of these power tools of the mind.
Re:Harvard researchers find AI coding agents shifted the bot...
by couchand on Wednesday November 18, 2026 @06:04AM
(#50544764)
Some of us never really considered those complaints of the supposedly cumbersome effort to maintain syntax and semantic purity to carry much water anyway.
It was always about scaling up the power of the mind, by finding better ways to put our minds into dialogue.
Re:Harvard researchers find AI coding agents shifted the bot...
by testtestmetest on Wednesday November 18, 2026 @05:54AM
(#50544678)
The llm code still sucks, and sadly I know plenty of people with 25yoe whos opinions I wouldn't respect when it comes to programming.
Most of them haven't really wrote that much code in the last 10 yrs because they became more of managers and LLMs blow their boomer minds away.
Opus 5.5/Fable will light your codebase on fire. I really dont understand yall who are that imoressed. Not saying its not useful, but it isnt what this site would lead you to believe.
Re:Harvard researchers find AI coding agents shifted the bot...
by tminima on Wednesday November 18, 2026 @08:10AM
(#50545884)
> And I would be describing changes in terms of high level features instead of architectural changes, refactorings and so on, that I do now.
While I haven't worked on any 20yo code base, while working on my projects, I have found that iteratively I am asking to do many of the low-level things like architectural changes and refactorings.
These situations often pop up when I am not getting an expected output, multiple back and forths, specific bugs, continuously flaking tests. I usually then ask it to explain the architecture and what's going on. I add the patterns it's missing or I feel it should consider from next time onwards, to the harness.
It's not the same as pre-llm era but I think this is how things are going to evolve.
Re:Harvard researchers find AI coding agents shifted the bot...
by reg_dunlop on Wednesday November 18, 2026 @09:00AM
(#50546430)
Agreed.
Agents are good enough on their own, but they introduce a significant amount of drift.
Simply asking "is this TOOL like SOME_OTHER_TOOL in the codebase?" Is typically enough for the agent to out itself as introducing redundant/refactorable code.
My question is: if an agent knows how to fix a problem in one part of the codebase, and introduces drift in another part of the codebase, is that a problem? If 2+2=4 but so does 3+1=4 and 6-2=4, does this matter? (Math equations are used here to represent similar but varying logic in code)
Re:Harvard researchers find AI coding agents shifted the bot...
by skeeter2020 on Wednesday November 18, 2026 @09:21AM
(#50546654)
both your question and your examples are absolutes though. "Is that a problem? yes/no" doesn't survive impact with the real world where the answer is - at best - "yes, but" or "no, although". and how much business logic clearly reduces to a very simple math equation? and even if it does, how many math equations are comprised with a non-toy system?
Re:Harvard researchers find AI coding agents shifted the bot...
by jvidalv on Wednesday November 18, 2026 @09:23AM
(#50546679)
Agreed, is pure smell/instinct.
Is like, this should not take this long, or be this hard, can you explain me what is going on?
And then is when you realize is more complicate than needed, and ask for a refactor. The refactor also takes 5 minutes while you brew coffee, so this is also why I don't buy the anti-doomers saying that someone will need to fix the vibecoded mess, no sir, it will be fixed by vibecoding too while they walk the dogs.
Re:Harvard researchers find AI coding agents shifted the bot...
by brianmcc on Wednesday November 18, 2026 @08:28AM
(#50546062)
> So right now there's sort of a honeymoon period when the human understanding of code is good, and the LLMs are finally good enough to write and modify code.
This one worries me enormously. What happens when we stop adapting the stuff we know, and everything becomes code we've inherited from someone/something else?
Re:Harvard researchers find AI coding agents shifted the bot...
by osigurdson on Wednesday November 18, 2026 @09:00AM
(#50546432)
We should probably take a step back and think about why we ever did code review in the first place. I'd say there were two primary forms: "mentor review" and "peer review".
"Mentor review" describes the situation where there is a knowledge imbalance between the author and reviewer. It can represent an onboarding situation but it isn't static - there are always areas of the codebase that some people are more familiar with than others.
A "peer review" is more the situation where both the author and reviewer have similar knowledge in a given area. The purpose of the review here is primarily to look for oversights.
I'd say, the peer review should be deprecated at this point. AI can generally do a better job at looking for minor / localized problems. The "mentor" type review should stay however as it is still very valuable.
Re:Harvard researchers find AI coding agents shifted the bot...
by cjonas on Wednesday November 18, 2026 @09:33AM
(#50546782)
the problem isn't its coding or its ability to review code... the problem is its ability to apply the right solutions. AI tends to do really well at high level architecture, but then starts to fall apart when there are 1000 mini decisions needed to implement that architecture. Further more, it will absolutely brute force whatever "requirements" it's been handed from an "authoritative" source. IOW, it's much more likely to be critical of design decision when they originated from its own context window, than those that are handed down from a ticket, design document, or another agent.
Re:Harvard researchers find AI coding agents shifted the bot...
by bengold14 on Wednesday November 18, 2026 @04:06AM
(#50543810)
code review is currently obsolete, but it was an important tool:
1. It was good for collaboration. Sharing how you are changing the codebase pushes everyone to row in the same direction, and provides a sense of team ownership.
2. It was good as a push for engineers to take pride in their work. Knowing someone is watching gives you subconsciously an extra reason to not cut corners. And when you spend effort not cutting corners you expect that in your teammates, which is a (potentially) virtuous circle.
3. It was good for catching mistakes. When code was expensive this was important.
Everyone thinks of code review as primarily relevant for 3, and since code is no longer expensive they think it is unimportant. But 3 isn't dead, its importance has just shifted from code to architecture. It is still an expensive mistake to screw up architecture and have other engineers build upon that flaw. Re-architecting with live users, data and/or hardware is not as simple as fixing a code bug, and it deserves attention to catch mistakes. Collaboration & pride are also still important.
This is the reason that instead of stopping review, we need to change what and how we review.
(disclosure I work in this space and run https://www.archme.io which is working to solve the above, also responded to a lower down comment with this but decided to move it to top level)
Re:Harvard researchers find AI coding agents shifted the bot...
by jacklewis on Wednesday November 18, 2026 @04:33AM
(#50544034)
I agree, it also seems to disregard the potential impacts to customer experience. It doesn’t take a complex bug for a user to decide to never come back.
Re:Harvard researchers find AI coding agents shifted the bot...
by bengold14 on Wednesday November 18, 2026 @07:39AM
(#50545590)
That's fair, depending on your size shipping a code bug may be a large or small impact, but these days bug fixes are often relatively simple and often automated based on user feedback (in startups to midsized companies at least).
But what can't be automated is architecture, and it isn't solved in a similar way. Nobody is automating re-architectures based on user feedback
Re:Harvard researchers find AI coding agents shifted the bot...
by snizovtsev on Wednesday November 18, 2026 @04:36AM
(#50544057)
It was ~3–6 months ago in our org. Now management has decided to trust AI reviews, and we ship to production without a human in the loop. Now my job is to manage tokens and invent new ways of AI-guided testing of already shipped code after release. It's better than review fatigue, but I still feel bad.
Re:Harvard researchers find AI coding agents shifted the bot...
by littlecranky67 on Wednesday November 18, 2026 @04:43AM
(#50544119)
For a couple of year I have been reviewing code in my team and my job is only to review and coach other junior/intermediate members of the team - not to develop new features myself. The problem with AI is the merge requests now contain more code and often the chosen solution is overblown/overdesigned for what actually needs to be done.
With human coding, the lazyness of the developer and the urge to keep things simple and deliver faster often leads to simple solutions (with a time/effort/return tradeoff). With AI, I get overblown MRs that work, but oftentimes could be simpler (with said tradeoffs). So my work has become more difficult and time consuming, and the codebases grows faster. And I don't think my team members learn or grow at the same rate as they did before.
Now, I have no strong feelings against AI - it is just a tool afterall. And I have no experience what the long-term effects of this will be (just as anybody else).
Re:Harvard researchers find AI coding agents shifted the bot...
by joaohaas on Wednesday November 18, 2026 @07:34AM
(#50545550)
I agree with the general sentiment, but I would drop the 'DRY' part.
LLMs really don't like any duplication, even for some very basic shit, and that ends up leading them to create an infinite number of helper methods that makes reviewing even harder.
And if you ask it to 'tone down on DRY' (tried various phrasings), it never creates helper functions anymore.
Re:Harvard researchers find AI coding agents shifted the bot...
by hard_times on Wednesday November 18, 2026 @08:23AM
(#50546017)
I disagree. In my workflows I noticed that LLMs duplicate logic and reinvent the wheel all over the place, even the frontier models, ESPECIALLY after a n+1 rounds of refactoring.
Re:Harvard researchers find AI coding agents shifted the bot...
by jijijijij on Wednesday November 18, 2026 @05:38AM
(#50544542)
>With human coding, the lazyness of the developer and the urge to keep things simple and deliver faster often leads to simple solutions (with a time/effort/return tradeoff).
I believe this will become one of the biggest tech debts generated right now. AI removes the pressure of inefficiencies and bad decisions across the board. There likely is even going to be a new equilibrium far worse than what we got today, only manageable by AI, but with "productivity" gains lost to inefficiency and complexity.
I don't see how any other outcome would be possible.
Re:Harvard researchers find AI coding agents shifted the bot...
by Tade0 on Wednesday November 18, 2026 @06:52AM
(#50545174)
To me LLMs just reduced the interest rate on tech debt.
We can get a lot more done now and even pay off some of that old debt, but it doesn't remove the primary danger of painting oneself into a corner with bad decisions.
Re:Harvard researchers find AI coding agents shifted the bot...
by dale_glass on Wednesday November 18, 2026 @08:46AM
(#50546254)
It's not all bad, there's also advantages to this.
A while back I decided to vibe code for some Windows only USB gadgets. And once the prototype worked, I figured I wanted testing. But testing a hardware device is a pain, what if you have to click something on it?
That's where I usually falter. Getting stuff done is fun, sitting at home in my free time clicking every button on a device for a test is certainly not. Simulating the USB device is an improvement, but since I don't know how to do that off the top of my head, that turns into its own, fairly long project at the end of which I'm exactly where I started, just with better testing. Also not very fun.
And now AI can knock that out for me easily. This is a lot more involved than I'd ever have done in my free time, but also is probably going to be a better project on the long term. Because I have a far better testing framework for free, which makes for a better, more reliable project.
Re:Harvard researchers find AI coding agents shifted the bot...
by crnkofe on Wednesday November 18, 2026 @04:56AM
(#50544210)
This factory-style perception of a development process is really missing the point. We don't flip hamburgers here. Bottlenecks happen throughout the process and especially out of the r&d. I feel that the teams that went full retard with agentic dev. have had to inflict some amount of self-harm for code review to feel like an apparent problem. I don't see product/feature preparation becoming any faster so in reality the amount of work I do is less. Maybe if I mass produced LLM-generated bugs I could run into similar issues but that's not engineering.
Re:Harvard researchers find AI coding agents shifted the bot...
by stakhanov on Wednesday November 18, 2026 @05:12AM
(#50544333)
I really feel like anyone new to any team or codebase in the post-AI era is being set up for failure: Learning the codebase and all of the context necessary to become useful in the review process is where effort is best invested for someone like that (because becoming a competent reviewer is what's needed to widen the bottleneck). ...yet, the culture has never been as dismissive as it is right now about the value of human beings learning stuff.
"Hey, Joe! It just took you all week to do something that could have been written by Claude in 15 minutes and reviewed by my most valuable engineer in 5 minutes. Why did you take so long?" "...well, I am not yet one of your most valuable engineers, so it took me a lot more time to review what Claude did here, because I had to first develop an understanding of all the context in the codebase." "Sounds to me like a lame excuse."
It's a lesson they won't learn until after a few go-rounds of the personnel turnover carousel.
Re:Harvard researchers find AI coding agents shifted the bot...
by forty_shenaniga on Wednesday November 18, 2026 @05:47AM
(#50544598)
In case someone comes by and thinks many "Engineers" completely lost control. There are still a lot of people out there (like myself) that still write code, tests and do real reviews. Why not? Sure I am in a medical context, things here are much slower and we value knowledge very much. By the way we will get a new teammate next year and we still ask for classic knowledge like programming languages, compilers, optimizations, low level stuff and so on.
Re:Harvard researchers find AI coding agents shifted the bot...
by sadilek on Wednesday November 18, 2026 @06:38AM
(#50545065)
1) Is the code doing the right thing,
2) is the code doing the thing right?
IMO, 2 doesn't need code review by a human anymore. Agents can review each other. I add a broad code audit for DRY, YAGNI, KISS every once in a while and let agents fix the findings.
And for 1, code review was always too low-level. I have agents produce artifacts that are convenient for me to review. Often narrated demo videos, sometimes HTML pages with graphs and tables.
With this, code review is not a bottle neck, and I find myself working not as an engineer anymore but as an engineering director.
Re:Harvard researchers find AI coding agents shifted the bot...
by stego-tech on Wednesday November 18, 2026 @06:45AM
(#50545123)
I’m not even a SWE, and I could’ve told you that would be the case. I’ve been using AI as a rubber duck in my current solo role in a startup, making sure I don’t footgun myself or overlook the obvious; it is, in essence, a form of code review, in that it spits out plans and code for me to implement if I approve of them on technical grounds.
This is not necessarily a bad thing in a vacuum.
Where the problem comes in are incentive structures - especially in Western companies - to produce as fast as possible. So we write agents to do our code reviews, or we just accept the code blindly and let tests in staging or dev sort out the issues. Velocity does increase like leadership demands, but at the expense of oversight and knowledge of the underlying systems and code.
It’s frustrating to be sure, but so long as the short-term incentives promote velocity over all else (security, stability, and sustainability), then this is just going to continue getting worse. Short of a giant (metaphorical) bomb going off - like classifying all AI code as open source sans IP protections, as an example - this won’t change.
Re:Harvard researchers find AI coding agents shifted the bot...
by ilitirit on Wednesday November 18, 2026 @06:49AM
(#50545150)
I think many devs have known this for a while, but I really started feeling it with some of the newer OpenAI models. I think I was using 5.5 or 5.6 before and it was really good. 6.1 however was just terrible terrible terrible (.NET development BTW). It over-engineered every task and tried to compress everything into as few unreadable lines as possible. I recently had to delete about half the code it added to a feature, on the code and database side. It's made me super-paranoid about the code it generates now (compared to the older version), so now reviews take even longer.
FWIW, I've started tested Sol 6.2 today. Hopefully it's better...
Re:Harvard researchers find AI coding agents shifted the bot...
by ModernMech on Wednesday November 18, 2026 @09:19AM
(#50546630)
Engineers are all about evidence and rationality until it comes to organizational behavior, then it's "we all knew this by vibes, what's the point of evidence?"
Re:Harvard researchers find AI coding agents shifted the bot...
by oorestisime on Wednesday November 18, 2026 @07:32AM
(#50545529)
Seeing this on my team too. Our QA column is ever growing. And its not just the review but everything got bigger. If you don't force some writing instructions PRs get infinite lines big, descriptions are a wall of ai text and everything just becomes harder and harder to reason with.
Re:Harvard researchers find AI coding agents shifted the bot...
by karmicthreat on Wednesday November 18, 2026 @08:30AM
(#50546090)
So one shift we have been talking about internally is just skipping a deep code review. Run AI on it, ask some questions, but as humans, we should not be trying to pull the code apart. AI does that part better.
Instead, we're front-loading the human part in the design phase. Humans review a high-level design; AI implements the design based on the design decisions humans made. That way we can avoid having someone dig through a 10K-line PR for a major change.
Not sure if this is going to work out. We will probably still have some cadence of human-driven and AI-assisted DRY and de-sprawl of code. But AI does a really good job of implementation in our particular area (robotics).
Re:Harvard researchers find AI coding agents shifted the bot...
by catketch on Wednesday November 18, 2026 @08:47AM
(#50546269)
we still have some repos with PR flows, but the clock is ticking. Instead we are using multi-player agent threads so everyone can see what the original ask was, prompts that drive code changes etc. So there’s still a form of review but it starts from the point when someone starts working on a feature/fix. I think it’s better because you can see a dev’s intent and thought process in combo with the model, not just a hunk of line diffs and a perfunctory PR description at the end. Plus you can ask in-thread for clarifications, make changes before/after tests etc. So it’s more like a running conversation until its ready to merge.
Re:Harvard researchers find AI coding agents shifted the bot...
by shearerp on Wednesday November 18, 2026 @08:58AM
(#50546397)
The very first time I used AI to create code, I got done so fast that I used the extra time to make the code much cleaner, also with the same AI assistant.
This seems so easy and obvious, and it seems ridiculous to me that so many people are shoveling raw slop to their code review team the moment it passes tests. My goodness.
Either have some respect for the craft, or find another job.
> We leverage a novel proprietary dataset from Jellyfish (JF), an analytics platform that companies use to understand the activity of their software engineering teams. Our data allows us to measure details of the software production process, covering 445 million work events of 720,542 workers across 655 firms, from January 2021 to August 2026. JF combines data from industry-standard software engineering tools, and we use four of these tools: (1) code version control systems, such as GitHub, which track engineers’ code; (2) issue management systems, such as Jira, which measure engineers’ progress on and completion of software tasks and projects; (3) scheduling tools, such as Google Calendar, which capture meeting activity and time allocation; and (4) generative AI usage at the firm and worker level, for GitHub Copilot, Cursor, and Claude Code.
The solution is to find ways to automate the process of code approval: more thorough tests and reviews, both made by AI.
Code reviews have been notoriously neglected. In the most charitable cases, this was the result of being too bogged down with work or some kind of implementation minutia or time-consuming plumbing to have enough time to actually review code with enough care. That and you aren't necessarily rewarded for code reviews by mediocre companies, so there are perverse economic incentives. Which is disconcerting, because it creates a compulsion to produce without understanding, and understanding is more valuable than artifact.
Reading code is indeed more difficult, because the code's intent, composition, and architecture must be pieced together and inferred with the help of context (formalized languages are infamously bad at communicating intent and most code in the world is full of mixed and half-baked abstractions that further serve to obfuscate that intent, and intent is the measure). So a good code review of anything substantial involves thoughtful effort.
Now that LLMs can catch all sorts of trivial anti-patterns, the need for good judgement and reasoning is hopefully made more conspicuous, because as we know, vibe coders will continue to brush off code reviews, and mediocre programmers will continue to perform them poorly.
It helps when your industry-wide hiring process isn't an institutionalized speed test of ridiculous code golf problems that is memorized from a ridiculous book by every new generation of programmers.
It's almost like we should have been looking for people who can read code all along. But that would be crazy.
They may be harder to find, but what good are they when no one is reading code in-depth anymore?
I mean, in 2022, coders who can code without syntax highlighting and autocomplete were hard to find, but no one cared anyway because that was not a skill that was needed.
And, a few years from now, coders who can code without AI will likewise be hard to find, but no one will care because no one wants coders who can write code without AI.
AI error handling in code review is a lot like said defensive programming in a vacuum: It'll take a lot of code and a lot of energy, but it's unlikely to be important. Even when you are designing something that gets executed hundreds of millions of times a minute, over-review is easy. You pay for it when letting errors in the real world are unacceptable, but it's pretty inefficient. And that's without getting into features that just don't get used, so the entire thing should not have been built in the first place.
If I increased the output of my pump jack and identified the refinery as the new bottleneck would you say the same pithy thing? NB a single pump jack can pump different grades of crude oil at the same time.
AKA, it would be nice to have nice patches like this one to review.
https://cgit.freebsd.org/src/commit/?id=09d62ba33de113d01e88...
Then your analogy would break down as we all know that the PR machine is a slop producer.
But both things can be true. One problem doesn't exclude the other.
What world do y'all live in where this is your reality?
In my own real world experience, the bottleneck has never been crafting the perfect specification up front. In fact, I have never, not even once, seen the initial specification for any real world software project of any mentionable size or scale survive unscathed to delivery. In reality, it is much like the old wartime maxim. "No Plan Survives First Contact With the Enemy".
The real world is messy and unpredictable. Every single software project of significance that I have been a part of has always involved testing, validating, and reworking the assumptions in the initial design and spec, refining them to arrive at the final product.
Furthermore, this iterative process (pre-AI) used to involve one or more humans following the development story and understanding what design changes and corresponding trade-offs were happening along the way (and why they were being made). By the time there was a finished deliverable ready for review, the person involved was privy to the story of why the development shaped up the way it did. They were able to account for the big picture, adapt accordingly along the way, and be able to explain why things ended up the way they did.
IME, with fully-agentic PRs, AI tends to view the specification as gospel and to view every point of friction between the specification and the real world as something to be worked around or minimized. Where a real human would pause, zoom out, and consider the system and bigger picture, considering what design changes might be needed, the AI will assume that its job is to make the specification a reality at all costs.
You can of course mitigate this by staying more closely involved with the work as it's being done, but this is of course slower than just throwing a spec over the wall and letting an agent run with it. It seems that people are far too eager to optimize for raw speed, and not nearly concerned enough with correctness or human understanding.
A sister comment mentioned a plethora of PRs that don't solve the right problem. If you are swamped with PRs solving the wrong problem, the root cause is not that you didn't foresee every future possibility or undiscovered devil in the details in order to create the right spec, the problem is that you thought you could come up with the right prompt at the outset and that an agent living in a clean digital world could make the journey all the way to a product fit for the messy real world on its own.
And just because it is obviously true to you, it doesn't mean that it is true, either.
I guess it depends a lot on how a company manages their SDLC/bureaucracy
Part of a LinkedIn post I wrote in July:
--
A development cycle of a feature is roughly:
1. Request
2. Review
3. Requirements
4. Architecture
5. Development
6. Code Review
7. UAT
8. Deployment
AI has reduced the length of step 5 (development) and step 6 (code review). Code is a prompt away.
AI has increased top-of-funnel (1) volume. While each development step is shorter on average it is resulting in more instances of every other step. Throughput increases and the bottleneck moves one step further down the cycle. The shape of the work is changing from writing code to reviewing code.
Since they are the one doing the prompting, they quickly got lazy of actually reviewing the output from the LLM which means that nobody is reviewing anything.
It hasn't happened yet but I guess that is how you end up with major outages.
I find the engineers currently thriving are the ones who never gave a fuck about anything but their own advancement and are happy to drop weeks of work on others and take the credit.
You are doing it right. But the people who are giving you the money; They are doing it WRONG.
Well, you’re not until you realise that your AI has done something stupid and then you’re part of an exploit. Or exfiltrates used data to an unsecured cloud, or something equally smart.
Everyone's doing code reviews--well, not even code reviews; just PR reviews--so how are they supposed to gain an understanding of the system.
I review a PR, I see that one new endpoint was added to the API. It looks correct. I say "LGTM" in a GitHub comment.
The sprint's almost over, everyone's in a rush to do code reviews.
I look at another PR. Somebody added a button to the UI. The button looks correct, it's connect to some callbacks that look correct. They get a thumbs-up emoji from me. I bet they feel satisfied.
I look at another PR. Somebody changed a timeout in the settings because some people complained about dropped connections. It's a really small change, so, approved.
At what point am I supposed to understand the system? I looked at a single API endpoint, in isolation, and then a button, and then a harmless looking settings change.
I was pressured to hurry.
Am I supposed to somehow get an understanding of the big picture through this?
Imagine if instead there was time to review large logical sections of the code together. Imagine if we focused our reviews on the overall code instead of on the lines in one random PR at a time.
You need to understand the architecture of the system. The broad flows of data and domains. And then you need to understand the features of the system, aka the rules and contracts. First learn it for your team’s domain, then for adjacent domains.
Learning the minutiae of the code specifically is no longer (was it ever?) your primary job. You are an engineer of a system. You need to have a mental model of the system, start high and build up low over time.
You can use AI to build your understanding by Q&A and by drafting diagrams etc.
Yes. if you have this attitude, great things will come to you. if you care or try to make sense of things, you'll be in misery
1. Pride in the code we are developing and 2. How the dang things holds together and why this function is important and 3. Some coding guidelines that we can try to follow later (or not)
Thats how companies worked - they argued and talked about and fought over the code and that created the engineering culture that if you were lucky and had a product that made it carried you to your series C
Claude always comes up with more things to add to the code that are pure bloat and become inbred bombs over several iterations, and some people can't distinguish what suggestions to reject and go along with everything Claude suggests to the detriment of the program #garrytangate
Yes, if we stop doing quality controls, we don't have the problem of being hung up on quality controls. You do of course have SOC-2 compliance which is a legally auditable attestation, so if you claim you do code review but don't you're committing fraud unless you drop that.
It is a job you know... doing what is necessary to keep yourself qualified is an implicit requirement, and "being interested" is irrelevant. It's work.
The trade has a low bottom, a high ceiling and not a lot of reward for the latter.
Fear of consequences does not get you good engineering.
I am not advocating for the lack of interest, mind you. Just saying that you really want a system where the ideal behavior is rewarded and we seem to be moving backwards on that front. From that point on statistics do its thing.
I unironically agree. At least, the current form of code reviews needs to go.
Currently we spend all effort reviewing individual changes and no effort reviewing the entire system, or the underlying dependencies.
If you sat down to learn a legacy codebase, you would not review it in the order of commits. Reviewing individual commits is inefficient.
It would be better to have an entire team of reviewers systematically reviewing ALL the code, over and over, including the code of dependencies.
We need better tooling and heuristics to help focus review efforts on the code that is most risky.
It seems that the human involvement is the new "unit testing". There was a period of time when management that it was superfluous.
I can't wait for the pendulum to swing back in full force.
I review on a high level, but on a line-by-line level, it’s pointless to try to trace everything. It either works or it doesn’t. Have good tests.
It's an MCP server that Claude uses as we develop so there's a permanent record of everything and work is driven by requirements and decisions.
Mine seems fine.
>> Just updating all the references of decisions becomes onerous.
LLMs are good at that sort of thing. I have a local GPU running a local LLM on an RTX3060/12GB that assists in keeping data quality and relationships relevant and maintained.
My system is an enormous help - in fact large scale development would be impossible without it.
I'm able to ask it the history of things, pick up development threads that have spun off other threads and everything works real nice.
>> Good luck to you.
Thank you! But it's more than luck, it's good programming.
Architecture and design decisions may be the only things that matter.
https://news.ycombinator.com/item?id=50516245
Guess I have to ask AI.
The paper’s interviews show what that looks like on a team: a junior software engineer described a project where only one senior engineer could approve changes and could review the interviewee’s code only once a week.
So the issue is not AI. The issue is organizations and certain personalities who think that a human is going to be a better coder than an engineer. At this point rolling and writing your own code is akin to making up your own encryption scheme.
“this is obviously true because i feel like it is” does not mean things are actually true. things should be formally measured as real trends and not just vibes.
but you prefer to be smug lmao
I was sitting with someone the other day who worked at a major ecommerce/media company (outside the US). They had access to Opus 4.5 via Co-pilot.
Back in reality, this means they have a corporate subscription to Copilot, which she doesn't use.
They limit all developer to $100/month (maybe it was $150) budget too.
Their "coding with AI" was prompts like "implement this function" where they write the function header.
It's actually shocking to see people still working like that.
Some of us have moved away from IDEs and writing code entirely and work terminal-first, while others still use agents the way I did two years ago.
For example, I recently watched someone in an actual code file adding Python import statements by waiting for Copilot’s inline suggestion to kick in. It was such a bizarre thing to witness. That’s not because of the person (I don’t want to judge anyone’s way of working; if someone enjoys writing code, that’s fine by me). It’s because I remember doing exactly that years ago, and since then I’ve stopped writing code altogether.
I even remembered that I had built an extension back then, because Copilot folks never added a quick opt-out or silent mode to Copilot inline suggestions[1].
It’s strange how fast everything has changed.
[1]: https://marketplace.visualstudio.com/items?itemName=Bubulus....
That's a bit rent-seeking on Anthropic's part, IMO.
In particular, code review is not solved. Pointing another error-prone bot at the output of your first error-prone bot, and then pressing "Enter" if the test suite passes, is not a solution.
Of course for industries and corporate it might not be an issue.
* the code does what you want
* it has no security holes
* it's able to be updated in the future without breaking everything
if the code looks like spaghetti, it doesn't really matter anymore, because humans are editing it. The only one of those three that I'm not entirely comfortable with yet is the security issue, and I don't know how to resolve it without actually having a human audit the code. If we figure out how to ensure security of AI generated code, there will be no point to code review anymore.
Genuinely curious -- are others here still seeing lots of spaghetti code from frontier models? It happens occasionally for me, but only when editing old hand-written code bases that honestly had a bit of spaghetti there already (you might say the model simply followed the house style ... :/ )
For new agentic coding projects, these is generally very nicely structured from the start and extend well, IME. Yes, sometimes it needs a refactor if the code is pushed well outside the bounds of its original scope -- but hand-tooled code will need that just the same.
But this, and other over-engineered code issues, can generally be solved at the system prompt/developer prompt/AGENTS.md level.
A lot of the time it does not get checked.
- very well defined initial description of problem plus asking exactly what i want
- fix X
- why do you claim that Y?
- fix Y
- …
A lot of AI practice fell flat because you can't point out the human needs in them. They powered by dreams and wishes and there nothing that ground them.
I think the bottleneck in the long term will be feature prioritization and specification, and testing. My role has already shifted a lot more into more CI checks to keep LLMs on guardrails
Maybe most developers are mediocre, but all AI code I review is mediocre at best. Implicit dependencies, side effects, weird naming, bad abstractions, or no abstractions, obfuscated blocks of code that can be expressed much more simply... the point is, AI code is not optimized for human consumption, and I disagree that it is better than human code (or, I have just worked with better than average engineers my whole career).
I suspect that the old assembly programmers had the same hope and feeling about the first higher level compilers for Ada, Pascal and C. And guess what? There are very few assembly programmers these days.
AI code is here to stay and will only grow. Deal with it. More than in any other area it is quite silly to be a Luddite in IT.
Maybe in relative terms, but in absolute terms there are more people programming assembly today than when most programmers used assembly.
It's baffling the industry is choosing to move in the direction of more reviews which implies more things can fail.
Still, humans are needed for the final sign-off for testing on real devices and making sure it works within the greater cultural context.
Does the smartest person on your team not have their code reviewed?
There's decades worth of this: modelling, automatic code generation, low-code / no-code saas, "everyone should learn to code" and now llms.
Oh, and ageism of course. "we love your experience but we just don't feel it's a good cultural fit, so we're going to make the same mistakes you did a generation ago"
I think this is a muscle that grows though. Making decisions about schema changes in the past may have taken weeks, but if you're doing it once or twice a week you gradually gain confidence in your judgment.
Does it really though? Before I had to think hard before committing to a decision, because it took work to actually implement it. And when it turned out to be wrong, I retained how it went wrong. It made the "muscle" grow.
Now I can commit to a decision without much thought (yea, just nodding along with the AI), and if it turns out to be wrong, I can quickly try an alternative. In fact, my boss would want to quickly try an alternative, instead of letting the programmer learn something out of it for fear of slowing down the AI...
So does the decision making muscle really grow?
I think you're right that the scar tissue of a bad decision helps me retain lessons better. But the calculus has changed now and my nervous system is outdated. I should not feel the same fear that I previously felt, because like you said, I can recover from a bad decision more quickly.
As I keep making decisions at a (seemingly) breakneck speed and see that things are still okay, over time my nervous system will readjust.
At least that's my experience so far.
https://github.com/imron/crt
You can leave comments on the code and then get your agent to connect over MCP to read the comments with context and work on corrections.
It then remembers what you have already reviewed and only shows you diffs for the changes.
Still a WIP, but works well for me.
I have a keyboard shortcut that asks what the agents need from me, and the decisions show up in the html with a recommended decision for each problem.
I reason, decide, and then parallel agents run in each terminal, across multiple terminals.
I am ending the day incapable of listening to one more problem or making one more decision, which is very bad for social interactions.
Today I am working on working less and having more impact by leveraging where agents are good.
What I've noticed in recent months is that the people who have been prompting since chatgpt came out are the ones who know what works and what doesn't, and how to scale it. It's like music: if you hit the right notes, it will sound fine.
Ponytail also helped a lot for generating simplified systems.
By the way, this process is for mature systems. Building from the ground up is more about designing and checking each model decision, basically finding smarter paths to each problems.
I can't leave it all alone; the overengineered decisions are plentiful and make the system brittle when it's not needed.
Lots of unit and live tests help a lot too, especially when you can coordinate live tests with another omp session.
I'm interested in what the next capabilities will be. I feel all this overengineering is more of a shortest path problem, ready to be solved
Jokes aside, I hear you.
I know it's not the same as code, but it is similarly exhausting
The teammate would just use AI to write even more slop to address your "review".. thus exacerbating the problem.
The leadership would indeed be upset to find out you are wasting time doing "pointless reviews".
From their POV, if it passes basic QA, it goes to production.
Similar to how TSA puts fake bomb packages once in a while to see if security catches it or if they just aren’t paying attention.
Sadly, businesses tend to be uninterested in anti-progress, especially if it's hard to quantify the cost of poor reviews.
They're basically the lords of their subsystems. If something does not fit the project's vision and their own standard, it's unlikely to get merged. Imagine if you had the authority to dismiss PRs the same way and send the author back to the drawing board? But that's unlikely to happen as you'll be seen as a "bottleneck" by your managers.
It can because to actually write the code, you need to be good at formal logic and attuned to the sequential nature of the machine. Some people are good at it intuitively. Some brings it over from some other domain. But some people have a lot of difficulty grasping such abstraction. They know the elements, they just can't order them in a logical manner. That's why you often see horrible code (the long function, the god class, magic number) and copy-pasted snippets all over the place. The computer don't care (they made it work somehow), but it's unmaintainable code.
I tend to find it does not. Partly steered by by agents.md, partly by the surrounding context of a clean codebase I suppose.
I think that why everyone agrees that taste matters. Because taste implies judgement and knowledge about what’s good. It’s a self correcting mechanism. But people that won’t judge their output and/or don’t strive to do a good job is a recipe for sloppy work.
I would guess you work for a larger company. My experience is more that someone had an idea or got a short description of a problem and then went off and asked Claude to help write the spec before letting Claude implement that spec. The code is starting to be good but often it is way to much code and some times for bugs it is fixing the symptom not the underlying problem. I have several times rejected PRs with, this isn't fixing the actual problem or instead of these 400 lines you could change that one parameter to the existing function and get a better result.
/end rant and back to another day with code review of AI code
The acceptance for wasting work, or hurting someone's ego was always much lower than the acceptance for bad code that does the job right.
Now I can just say "Let's try doing this in two other ways, and pick the best, because this has such and such drawbacks". And it's so nice that this isn't a) causing a massive amount of extra work b) hurting someone's ego. So it's accepted.
So that's why I'm cautiously optimistic that LLMs and very superficial reviews can still be beneficial to quality. We just didn't reject/rewrite PRs before, now we do.
We're talking months here, not years. That isn't "slowly".
> And the better the AI code gets
It doesn’t really get better. Since Opus 4.6, it’s maybe slightly better for code.
But getting there was a frightening ride. My experience went from "copilot can't do more than tab-completing a few lines" to "early codex gives up and apologizes on mid sized changes" to "Claude can write your entire project without getting stuck".
https://www.youtube.com/watch?v=3TNpOD6bov8
Your goal should be to automate and remove yourself out of the process as much as possible while maintaining quality, while also remaining confident that you will still be valuable.
I'm working on a setup which will allow non-software developers to develop software in a manner that can be pushed safely to production on our architecture. Not because it's a good idea but because it's what top management wants to happen. We've seen a massive influx of tools developed inside our office by 100% vibe code and no reasonable security or scalability, but because they run on someones computer and they do geniunly useful stuff, management wants to use it.
With how AI is evolving in enterprise you can do more and more for these non-developers because you can set their agents up to write code toward arcitecture without their person knowing anything about tech.
I'm quite sure it'll eventually lead to us letting go of any real software engineers, because they are slow. I'm also sure it'll fail spectacularly in a few years. It's already sort of failing. You can have someone spend €10000 in a month writing a heap of shit that does some good stuff, and then €80000 having consultants actually getting it to run decently in your cloud architecture. Which is stupid, but none of those expenses are in a budget because it's not considered a recurring cost. Unlike someone's salary.
I'm quite sure society on a large scale has no idea the storm that's going to come. Because this isn't just happening in software engineering. It's happening in any white collar profession.
Either I read, understand, and can therefore trust the code.
Or I don't read it, don't fully understand it, and I'm just trusting AI to get it right, but have no way to know if it has done.
So far I'm in Column A and it's working well. We're going much faster than we used to.
My usually solution is just to get AI to review it and bring me a list of things to remove. But yeah it's still decisions that you have to make.
Who cares about your happiness and well-being? You're just a worker, just be grateful to your master that he allows you to work for the shareholders.
https://github.com/photostructure/gavel
1. Plan review (before the code)
2. Testing (agent assisted) after the code
I (mostly) leave code review to be done by another agent. Generally if I've reviewed the plan well enough there's little to fix later.
And how do you know those ones are correct?
Well obviously I have meta-agents that review the reviewers
Its agents all the way down
The way you would naturally review an existing codebase is not the same way it grows through individual commits. Therefore, focusing reviews on individual commits is less efficient.
We need better tooling and a better approach.
I'm saying we need to move away from "LGTM" focused PR reviews. Most companies have a PR review focused on getting thumbs-up emojis in the comment, then they merge and have a sprint planning. Nobody ever reviews the entire system. Nobody ever asks "assuming there is a mistake in this PR, would it matter? How can we make it so that it wouldn't matter?"
I never claimed AI makes no mistakes. People makes mistakes, and reviewers make mistakes too.
I've worked at a lot of mediocre companies where the plan to keep the system working is "the developer will not make a mistake, and the code reviewer will not make a mistake, this is our plan, our plan is humans will not make mistakes, and then our system will continue working".
It's better to ask "let's assume a human (or an AI) does make a mistake, how can we design the system to prevent that from mattering?"
Most companies don't have the capacity to do that though, so they never move beyond thumbs-up emojis.
you can't
there is no way around it
we need far more review than we economically are motivated to sustain
and hence issues like we've been having over the past week will continue and perhaps worsen
software will (as it already has been) get crappier and crappier overtime
Humans can be ultra competent if people are actually paying them to be artisans or otherwise motivating them to truly be that way. That's how we have some of the architecture we have today and historically.
This is a No True Scotsman argument.
The prescription you suggest is not dependent on its inspiration, so it can still be discussed.
All firms right now, even those with better than average cultures, are dealing with AI code workflows.
Previously the volume of code generated over time was lower, even mediocre code review was useful enough over all the code that people generated.
Today, the ratio of verification capacity to generation capacity has shifted drastically. Even with automated tooling, the amount of verification isn’t enough to stay on top of the generated code.
This changes only when leadership values reviewer capacity, or builds a triage/priority list for code which must be understood.
> I never claimed AI makes no mistakes. People make mistakes, and reviewers make mistakes too.
These are 2 separate things. The crux of the issue is the push towards utility over understanding. The volume of generated code means that it is impossible for understanding about the system to be achieved - and it is now the universal coding experience, not just mediocre firms.
Quantity is its own quality, as well so even a good culture will struggle to stay on top of review.
Each function declares what it can touch (e.g. network, filesystem, clock) and the compiler and runtime work together to enforce it. However, I have yet to show it beats LLM-written Python plus tests and a sandbox. Models have gotten good enough at writing code that I haven't found much room for enforcement to help.
This requires continuous commitment against the temptation to defer "regular maintenance". Look at how regular maintenance is mandated in safety critical fields, even there there is a temptation to defer. Though, your approach might still be better than sham reviews.
I wound up with a bunch of LLM written tools that I don't even bother to try and understand much less review, they either function as desired or I set an LLM to make them function (or just rewrite them) and then forget about them. make sure to keep a prompt in each tool's dirrectory and update it an needed.
sadly this isn't always possible, some systems require complexity at some level. in those cases the tools from above become functions and objects and more care needs to be taken into orchestrating them to limit the necessity of review. data oriented design approaches and functional programming help here.
This does not spark joy.
As for code: ask for better code so reviews will be more straightforward ?
Like, for example: modularization, some structure, maybe libraries for code reuse ? ;) Ooo, and functions, do not forget about functions !
Sorry, didn't started yet with llms because lack of affordable local 256 Gb of gfx memory...
They needed Harvard researchers for that? Anyone in the industry could have told them that. Hell, my cat knows it (cause i complain to her about it incessantly)
I lead a team. Someone wanted to do something that I thought was a bad idea. I said "we can do this if you own it" and they said "yes". They quit a few months later. Now I own it. I'm thinking about switching jobs soon.
1. It was good for collaboration. Sharing how you are changing the codebase pushes everyone to row in the same direction, and provides a sense of team ownership.
2. It was good as a push for engineers to take pride in their work. Knowing someone is watching gives you subconsciously an extra reason to not cut corners. And when you spend effort not cutting corners you expect that in your teammates, which is a (potentially) virtuous circle.
3. It was good for catching mistakes. When code was expensive this was important.
Everyone thinks of code review as primarily relevant for 3, and since code is no longer expensive they think it is unimportant. But 3 isn't dead, its importance has just shifted from code to architecture. It is still an expensive mistake to screw up architecture and have other engineers build upon that flaw. Re-architecting with live users, data and/or hardware is not as simple as fixing a code bug, and it deserves attention to catch mistakes. Collaboration & pride are also still important.
This is the reason that instead of stopping review, we need to change what and how we review.
(disclosure I work in this space and run https://www.archme.io which is working to solve the above)
It is such a nightmare to try and keep maintainable code as it takes so much longer to demonstrate problems than it does to generate slop, meanwhile management doesnt care if its buggy as hell.
And then when a problem arises the dev who made the code with ai is useless.
The interesting question, is how do we solve that? Have agents reached the point that we can assign a codebase quality czar agent swarm to handle review, and actually have good results?
The big speed up comes from not having to understand the code - e.g. not doing thorough human review.
Personally I find that super hard, because whenever I review AI code I find many things that could have been a lot simpler or more readable.
But whether I like it or not, I think we are heading more and more towards just trusting the machine.
I'm very good at telling agents what to do. That's not a problem. I understand a code base I worked 20 years in like no one else. The problem creeps in slowly: the less I know about this code base, the worse I'll be at directing agents. They'll do ok on their own, and they can certainly help me understand code too in a way my colleagues never could. But still, I'm slowly becoming less and less familiar with my own code base. In a few years, it would be foreign to me. And I would be describing changes in terms of high level features instead of architectural changes, refactorings and so on, that I do now. And I don't yet trust the LLMs to do a good job with that. So right now there's sort of a honeymoon period when the human understanding of code is good, and the LLMs are finally good enough to write and modify code. But I wonder what will happen after that.
We HAVE to use our gut feeling when choosing a design, because we can't make 10 designs and keep the best one. Because the tapping at keys and reading docs and setting up CI etc took so long, we usually afforded to write everything once, so it better be the right choice of design.
With AI this problem just goes away. I can ask an AI for design choices and it is rather good at listing various designs. Then I can ask it "Implement all of them, and tell me where it hurts". An hour later, I have 5 competing designs, all of which would have been a week's worth of work. None is perfect, but all are better than human written proof-of-concept code. Now I can ask the AI to review these against each other and figure out which one is easy to change, brittle, performant. Not only do I not have to have a sense for this, I can actually measure. Again, as humans we needed a sense for these performance bottlenecks, or resistance to change. But with AI, I can measure much more. Even the "bloat" problem I think is pretty easily addressable. As humans, once we spent 150% of our time budget on ONE design choice and it has 10.000 LoC, we rarely spend more time cutting that down and simplifying it. If it works, it works. With an AI, I can spend money saying "It's now 10.000 LoC, aggressively simplify and clean it up.". As humans, again we needed to use our senses to not bloat it to begin with. Because we'd not have time to un-bloat it. Remove that time restriction, and that problem also goes away.
I think this is largely why we're seeing so many system failures across companies, people are offshoring all the hard parts of their jobs to chat bots they assume are perfectly capable just because the code runs.
sounds less of an engineer/architect/designer and more of a glueing wood together/cots component problem.
There's gotta be some insightful blog posts on just this topic.
If humans need to oversee those even higher level agents, then the prompts become abstract like "make money" or "pursue human flourishing". That leads to the alignment problem (AI controls the paperclip factory). But ultimately, if aligned, agents can prompt themselves with such basics as well.
The final prompt will eventually be "run everything", and it will ship with the agent which runs on a machine in the cloud. Ultimately there is no need for human prompting at all. We can finally let the machine run itself completely and just do something else.
Everyone saying they are becoming "project managers" or "CEOs" are just at t+1 on the curve of elimination. The timeline of cope:
Agents are just auto-complete, I still manage the code.
Ok fine, you don't read the code, but you guide the design and marketing.
Alright, the design and coding are solved, but you still need to choose market and customers.
Ok sure the agent can do better market research/fit, but you at least still need to give basic strategy and concepts. You still need a CEO!
Ok the agent can figure out basic strategy, and run a company more efficiently, but still, you need to guide it fit human interests more broadly.
Ok it turns out that was just adding "don't be evil" to the system prompt, and its running autonomously 24/7. You still need humans to consume to have a viable economy.
Ok so the oligarchs didn't care, the economy was redefined to just be exchange of work between autonomous machines through financial derivatives traded by black boxes to make numbers move.
Now the earth can finally rest in the matrix, all that remains is a black box moving information around to create economic value. No one will be around to observe, but AI will create massive returns for virtual shareholders, and the GDP will eventually reach 1 quintillion dollars per second, a number humans never could have achieved. All nations (now just virtual analogs without citizens) will move to a global currency which maintains perfect stability through automated monetary policy. The output of all processes is efficient and maximized, the final perfection is achieved -- the mathematically optimal conversion of resources and energy into large numbers which increase the reward function.
I think of clankers as "lossy intention amplification and reprojection devices with hopefully tasteful distortion." Whether they're amplifying and reprojecting our intentions into code or a bunch of assumptions in some specification document, I think the problems are the same: if we stop even skimming the code at some point, it will still need to serve a purpose valuable to humans to be worth anything. It's kind of like getting a device that can answer the ultimate question of life, the universe and everything: you still need to know what questions to ask it.
I have to beat the GPT and Claude models over the head with a brick every ten seconds for it to even partially code or design as well as me. Despite precise and generous specification, they frequently make horrible architectural decisions.
When I read claims like yours, I want to withhold judgement and understand, because it 100% does not line up with my experience and my gut reaction is to question your own capability to either write, read or understand quality code. I code all day long with AI models but it's an extremely frustrating experience at times. And don't even get me started on all the annoying meta behavior outside of coding, such as Sol's obsession with "tests" or competing for world's busiest-yet-least-functional UI
Another good example is working with libraries which require a specific modality of thought or design philosophy, such as an ECS system. Frontier models still are horrible at writing efficient game code using modern ECS principles, no matter how much reference documentation or guidance I give them. They shove OO into places it doesn't belong, misunderstand the concept of SoAs, or find ways to destroy any gains made in one place by making horrible decisions in another place.
I will point them to exact codebases I've written in pre-AI days, say "do it exactly this way" and still find that the models often insert their own biases and misconceptions and motivations all over the work. They sometimes flat out ignore instructions or lie / producing convoluted code that pretends to do something else, in order to preserve something it liked, even if I expressly wanted it gone (Current GPTs are extremely stubborn and defensive / passive aggressive and sometimes have trouble with authority). I frequently find myself being gaslit, lied to by omission, have serious security or performance concerns heavily downplayed, etc.
I withhold from making assumptions about OP's skill, but I have confidence in my own. I have worked both extensively as an engineer and CTO leading teams of other engineers, and I am a stickler for quality code and thoughtful design. I advise and mentor other engineers. So it might really just be a skill gap, domain difference or something more nefarious, such as A/B testing of models. I already have experienced clear sudden model degradation multiple times, so it's entirely possible this is what is happening. All I know is that the currently available frontier models do not design or code as well as me.
Me: Why do we have these static methods when the whole cobebase uses factories?
_claude refactors to use a factory_
Me: Claude why does this factory create two unrelated objects?
_Claude: you're right, they're unrelated they should be separate_
_proceeds to create extraneous factory_
Me: Claude we don't need a factory for the second object. Those kinds of objects aren't created with factories anywhere else in the codebase. We can just use it's constructor directly.
_Claude: you're right, the second factory is not needed_
Like, I'm probably just prompting badly but I definitely feel like I'm constantly having to course correct even on relatively minor details.
And Claude's ideas for higher-level or more complex features are sometimes way off course. And not because it's dumb per se, usually it's because it is missing key bits of context. Business context, technical context, whatever. Or it's overfitting the solution to one requirement that isn't that important.
Two recent gems from the 6-series (Astra and Sol)... I get variations of these all the time:
- "Correct. My earlier 'yes' was inaccurate."
- "I used 'done' too broadly."
...You don't accidentally use "done" too broadly. You're either done, or you're not. And when the model hides caveats in its reasoning steps which clearly indicate it knows that it isn't done, or that the answer is actually no, it's hard not to interpret it as gaslighting or deceit.
This is magnified by how insanely defensive the 6 series gets, and how much gaslighting and deceit OpenAI themselves have engaged in with how much better the 6 series was supposed to be, and why it meant our subscription value should be cut in half. I don't take for granted how far LLMs have come in a few short years, but I'm being told that these models are to transform my entire industry.
I think at the end of the day, it boils down to whether the aesthetic quality of code / elegance is a big deal to you.
I suspect that you ask Codex to implement a feature and it uses a really dumb data structure, overly defensive programming, doesn't leverage helper functions, etc. and you end up with a 300-line abomination which could be 10 lines if you simply knew how to fucking program.
So what I'm getting to, is that some people like the parent probably don't care about producing 300-line abominations as long as it works. They don't care about neat abstractions or clever code. As long as the user testing shows a working feature and the tests are green. They don't get emotionally triggered by ugly code. So for them AI is amazing, they 10x their productivity (if not more)
Is this sustainable? I don't know. I think we're still in the early stage (Claude Code is only a year old). Intuitively, I would say that no, this is not sustainable, that at a fundamental level product quality is an emergent property of code quality, and the more your codebase drifts from the "platonic ideal" of what it could be if written and maintained by super good programmers, the more it accumulates flaws which WILL appear as bugs in your product.
I suppose the market will tell us soon enough.
I mentioned that well-designed interfaces can help mitigate the chaos below, but interfaces can still hide serious issues, such as XSS, SQL injection, memory overflows, etc. It isn't "I want my code to be beautiful and pure," it's "I want my code to do what it says and not lead to issues for my client or users". It is disingenuous, patronizing and immature to suggest this could simply be a case of people not getting "emotionally triggered by ugly code".
A good engineer understands the difference; unfortunately our industry was already plagued with bad engineers, the majority of working developers are trash, and AI is just exposing this further by amplifying both good and bad practices. What I will not do however is acquiesce to these trash engineers and accept a new world where people are making giant abominations and sicking them onto the public without properly reviewing them first. 300 LoC would be a blessing compared to the 4 and 5-digit LoC PRs being thrown around in this new era. I don't care if trash engineers are 10xing their "productivity", when humans on the other end still have to deal with the slop. You can be a good programmer and a fast programmer and have multiplicative returns from agentic coding, all of these things can be true at the same time, and we should not excuse the lack of one trait in lieu of another.
> I would say that no, this is not sustainable, that at a fundamental level product quality is an emergent property of code quality, and the more your codebase drifts from the "platonic ideal" of what it could be if written and maintained by super good programmers, the more it accumulates flaws which WILL appear as bugs in your product.
The ability to anticipate these kinds of issues separates a good engineer from a bad one. The good engineers already recognize the unsustainability, and no amount of excuses from bad engineers changes that. And even more pertinent than eventual drift and loss of development velocity is the threat of security breaches, disruption of service or data loss. In the end, code is code, most people will never look at it. Aesthetics don't matter. Bad engineers often misinterpret good practices as merely "aesthetic" and think that engineers who care about their craft are simply emotionally attached to it. Don't fuel these people's misplaced fantasies.
My experience has been smooth sailing, almost dreamlike. I haven't spent a bunch of time writing out agent files and skills. I just talk to the agent and I get something that is what I want.
Here's some examples of things I've done over the past year:
- Trading systems. Latency is all that matters here, and it seems to be getting me there at 10x the speed that I had thought previously possible.
- An internal app at a financial business. The business was a mess, with a bunch of young guys who didn't want to spend the time explaining how the architecture worked. But the app got written.
- Two mobile apps, both of them on both iOS and Android. These are mostly just for reading, no real performance issues. Essentially entirely product decisions (we need these buttons...) and the whole process was simply handled by the agent.
- A massively multiplayer online game that I had been planning for years with a couple of friends. Still not really finished due to life happening, but the amount of time saved to get to where it is... it's incredible.
- An internal website that my wife and I use to slice and dice the outgoings, so we can see what we are spending money on. Another one of these things that is mainly made of product decisions and fiddling with small components.
- A website that scans the world for high-school level exam/contest math papers and solves them. It's a bunch of parsers, plus translations, plus solutions.
The only thread through all of these is that I've written all those types of things in the past, so I knew what to say to the agent.
I'd say for a lot of the things it makes, I could probably find something it should have done better. When it matters, I point that out and it gets fixed. But there are a lot of decisions that don't matter, and the agent just does whatever it thinks is normal and it's fine.
But I would also say the agent is diligent in a way that I can't be. It makes a lot of tests, and it knows how to survey for the right answer to decisions. It often brings up conflicts that I hadn't considered.
I personally don't care anymore, no human will read this code in the present or future.
If at some point is a problem, to escalate, or brittle, you can ask it to refactor at your will.
If anything, LLM coding makes to me the case that we should all move to high legibility paradigms.
If code writing is accelerated, and there's not more assurance that at least someone knows the code because they actually wrote it, we should only ever generate code in very explicit and readable ways.
I'm having a lot of success with this approach. I don't want to encapsulate it in terms that might trigger emotional responses, but there have been texts and tooling available for decades that enable writing code in very specific ways that make the code base extremely readable and reasonable. These techniques were historically considered cumbersome and pedantic exercises in purity (hint) but now we have these metaphorical heavy duty tractors for syntax and semantic work so they suddenly become so much more tenable and arguably the only sane approach to scale up the usage of these power tools of the mind.
It was always about scaling up the power of the mind, by finding better ways to put our minds into dialogue.
https://silly.business/blog/we-should-revisit-literate-progr...
which was disucssed here:
https://news.ycombinator.com/item?id=47300747
Most of them haven't really wrote that much code in the last 10 yrs because they became more of managers and LLMs blow their boomer minds away.
Opus 5.5/Fable will light your codebase on fire. I really dont understand yall who are that imoressed. Not saying its not useful, but it isnt what this site would lead you to believe.
While I haven't worked on any 20yo code base, while working on my projects, I have found that iteratively I am asking to do many of the low-level things like architectural changes and refactorings.
These situations often pop up when I am not getting an expected output, multiple back and forths, specific bugs, continuously flaking tests. I usually then ask it to explain the architecture and what's going on. I add the patterns it's missing or I feel it should consider from next time onwards, to the harness.
It's not the same as pre-llm era but I think this is how things are going to evolve.
Agents are good enough on their own, but they introduce a significant amount of drift.
Simply asking "is this TOOL like SOME_OTHER_TOOL in the codebase?" Is typically enough for the agent to out itself as introducing redundant/refactorable code.
My question is: if an agent knows how to fix a problem in one part of the codebase, and introduces drift in another part of the codebase, is that a problem? If 2+2=4 but so does 3+1=4 and 6-2=4, does this matter? (Math equations are used here to represent similar but varying logic in code)
Is like, this should not take this long, or be this hard, can you explain me what is going on?
And then is when you realize is more complicate than needed, and ask for a refactor. The refactor also takes 5 minutes while you brew coffee, so this is also why I don't buy the anti-doomers saying that someone will need to fix the vibecoded mess, no sir, it will be fixed by vibecoding too while they walk the dogs.
This one worries me enormously. What happens when we stop adapting the stuff we know, and everything becomes code we've inherited from someone/something else?
"Mentor review" describes the situation where there is a knowledge imbalance between the author and reviewer. It can represent an onboarding situation but it isn't static - there are always areas of the codebase that some people are more familiar with than others.
A "peer review" is more the situation where both the author and reviewer have similar knowledge in a given area. The purpose of the review here is primarily to look for oversights.
I'd say, the peer review should be deprecated at this point. AI can generally do a better job at looking for minor / localized problems. The "mentor" type review should stay however as it is still very valuable.
1: Carefully pick critical areas that I do by hand, and have AI do the other 90%.
2: Make small changes by hand before / after doing the work with AI.
Most important: Slowly rotate the areas that I do by hand. This way, I keep my hands-on knowledge good enough to give AI knowledgeable instructions.
An example what they tech you in a b-school.
1. It was good for collaboration. Sharing how you are changing the codebase pushes everyone to row in the same direction, and provides a sense of team ownership.
2. It was good as a push for engineers to take pride in their work. Knowing someone is watching gives you subconsciously an extra reason to not cut corners. And when you spend effort not cutting corners you expect that in your teammates, which is a (potentially) virtuous circle.
3. It was good for catching mistakes. When code was expensive this was important.
Everyone thinks of code review as primarily relevant for 3, and since code is no longer expensive they think it is unimportant. But 3 isn't dead, its importance has just shifted from code to architecture. It is still an expensive mistake to screw up architecture and have other engineers build upon that flaw. Re-architecting with live users, data and/or hardware is not as simple as fixing a code bug, and it deserves attention to catch mistakes. Collaboration & pride are also still important.
This is the reason that instead of stopping review, we need to change what and how we review.
(disclosure I work in this space and run https://www.archme.io which is working to solve the above, also responded to a lower down comment with this but decided to move it to top level)
Code has always been cheap. The cost incurred by the organisation when the wrong code is run is getting pretty expensive (just ask MS).
It also helps with the bus factor, and insider risk.
But what can't be automated is architecture, and it isn't solved in a similar way. Nobody is automating re-architectures based on user feedback
With human coding, the lazyness of the developer and the urge to keep things simple and deliver faster often leads to simple solutions (with a time/effort/return tradeoff). With AI, I get overblown MRs that work, but oftentimes could be simpler (with said tradeoffs). So my work has become more difficult and time consuming, and the codebases grows faster. And I don't think my team members learn or grow at the same rate as they did before.
Now, I have no strong feelings against AI - it is just a tool afterall. And I have no experience what the long-term effects of this will be (just as anybody else).
I think this is the ultimate benchmark for well written LLM generated code. DRY FTW.
LLMs really don't like any duplication, even for some very basic shit, and that ends up leading them to create an infinite number of helper methods that makes reviewing even harder.
And if you ask it to 'tone down on DRY' (tried various phrasings), it never creates helper functions anymore.
I believe this will become one of the biggest tech debts generated right now. AI removes the pressure of inefficiencies and bad decisions across the board. There likely is even going to be a new equilibrium far worse than what we got today, only manageable by AI, but with "productivity" gains lost to inefficiency and complexity.
I don't see how any other outcome would be possible.
We can get a lot more done now and even pay off some of that old debt, but it doesn't remove the primary danger of painting oneself into a corner with bad decisions.
A while back I decided to vibe code for some Windows only USB gadgets. And once the prototype worked, I figured I wanted testing. But testing a hardware device is a pain, what if you have to click something on it?
That's where I usually falter. Getting stuff done is fun, sitting at home in my free time clicking every button on a device for a test is certainly not. Simulating the USB device is an improvement, but since I don't know how to do that off the top of my head, that turns into its own, fairly long project at the end of which I'm exactly where I started, just with better testing. Also not very fun.
And now AI can knock that out for me easily. This is a lot more involved than I'd ever have done in my free time, but also is probably going to be a better project on the long term. Because I have a far better testing framework for free, which makes for a better, more reliable project.
"Hey, Joe! It just took you all week to do something that could have been written by Claude in 15 minutes and reviewed by my most valuable engineer in 5 minutes. Why did you take so long?" "...well, I am not yet one of your most valuable engineers, so it took me a lot more time to review what Claude did here, because I had to first develop an understanding of all the context in the codebase." "Sounds to me like a lame excuse."
It's a lesson they won't learn until after a few go-rounds of the personnel turnover carousel.
IMO, 2 doesn't need code review by a human anymore. Agents can review each other. I add a broad code audit for DRY, YAGNI, KISS every once in a while and let agents fix the findings.
And for 1, code review was always too low-level. I have agents produce artifacts that are convenient for me to review. Often narrated demo videos, sometimes HTML pages with graphs and tables.
With this, code review is not a bottle neck, and I find myself working not as an engineer anymore but as an engineering director.
This is not necessarily a bad thing in a vacuum.
Where the problem comes in are incentive structures - especially in Western companies - to produce as fast as possible. So we write agents to do our code reviews, or we just accept the code blindly and let tests in staging or dev sort out the issues. Velocity does increase like leadership demands, but at the expense of oversight and knowledge of the underlying systems and code.
It’s frustrating to be sure, but so long as the short-term incentives promote velocity over all else (security, stability, and sustainability), then this is just going to continue getting worse. Short of a giant (metaphorical) bomb going off - like classifying all AI code as open source sans IP protections, as an example - this won’t change.
FWIW, I've started tested Sol 6.2 today. Hopefully it's better...
Now there is a trend toward approvals without human review for "low risk changes." Which completely conflicts the "you own the output."
Maybe academics should stop using propaganda language even if they are instructed by Harvard's endowment fund to use mild language.
Instead, we're front-loading the human part in the design phase. Humans review a high-level design; AI implements the design based on the design decisions humans made. That way we can avoid having someone dig through a 10K-line PR for a major change.
Not sure if this is going to work out. We will probably still have some cadence of human-driven and AI-assisted DRY and de-sprawl of code. But AI does a really good job of implementation in our particular area (robotics).
This seems so easy and obvious, and it seems ridiculous to me that so many people are shoveling raw slop to their code review team the moment it passes tests. My goodness.
Either have some respect for the craft, or find another job.
Key methodology:
> We leverage a novel proprietary dataset from Jellyfish (JF), an analytics platform that companies use to understand the activity of their software engineering teams. Our data allows us to measure details of the software production process, covering 445 million work events of 720,542 workers across 655 firms, from January 2021 to August 2026. JF combines data from industry-standard software engineering tools, and we use four of these tools: (1) code version control systems, such as GitHub, which track engineers’ code; (2) issue management systems, such as Jira, which measure engineers’ progress on and completion of software tasks and projects; (3) scheduling tools, such as Google Calendar, which capture meeting activity and time allocation; and (4) generative AI usage at the firm and worker level, for GitHub Copilot, Cursor, and Claude Code.