It's never been easier to replace chunks of code with sane software patterns, but you have to have a feel for those patterns. And also understand what's under the hood.
You folks speak like the only function of the agent is to spit code and features. Get a grip and treat your deliverables with care, otherwise you only have yourself to blame, not the AI.
[0]: https://hn.algolia.com/?dateRange=all&page=0&prefix=false&qu...
You'll end with spaghetti if you'll play a bad manager and only ever allocate time for new features and never for cleanups.
You can go through code, add REFACTOR comments based on your tastes and thoughts, and get your result and iterate to your heart's wishes. You just don't need to do the direct code typing.
More broadly, it's well understood that experiments are not a replacement for design and UX. Google is famously great at the former and terrible at the latter. Sure the AI maxxers will say the machines are coming for all creative endeavours as well, but I'm going to need more evidence. So far, everything good I've seen come from AI still had a human at the wheel, and I don't see that changing any time soon.
“Play” is what produced at least two (likely more) generations of attentive (and therefore competent) programmers. The hype around LLMs is painful, yes, but attentive human minds will ultimately bust through it.
And then there's the times when the quick sloppy poc you planned to throw away gets forced into production and is still impossible to change ten years down the road.
AI makes all these problems so much less painful.
I worked at a company which had a huge monolithic ERP system (their product, to be clear) with no good separation between the GUI layer and presentation layer. The GUI was also dependent on an ancient version of the Borland C++ compiler. They put in a humongous effort to move to a slightly more modern UI library, and a client server architecture.
However, someone had decided that messages in xml or json were too inefficient, they already had performance issues. So they went with a binary message protocol of their own design - with no features for protocol update. Everything communicating with the server had to be on exactly the same version, or it would throw an error. So of course they very, very rarely updated the protocol.
I think the best help of AI will be to clean up such real life messes of soul-crushing architectural regrets. Will it do it perfectly, certainly not, but I wouldn't do it perfectly myself either if I was forced to do it - and I'd take a hell of a lot more time to do it.
No, because no amount of experimentation can solve many of the problems that have been solved by thinking. Even your claim about "experiments are cheap" requires thinking to decide what experiments to do. No one is generating all possible solutions that fit in X megabytes; you have to think to constrain the solution space.
We saw a similar philosophy in TDD advocacy many years ago. Search for something like "Sudoku Jeffries" to see how that went. Then search for "Sudoku Norvig" to see what it looks like when you actually understand the problem.
The idea that you can somehow iterate your way to a solution when you have no idea where you're trying to go or even which direction your next step should be in has always seemed absurd to some of us but in the era of LLMs there's no longer any doubt. In the agentic era (can we call a few months an "era"?) I estimate that 90% or more of the writing I've read about how to use agents most effectively came down to making sure there is a clear specification for what they need to implement first and then imposing extensive guard rails to make sure their output does in fact follow that specification. It's all about doing enough design work up front to remove any ambiguity before coding the next part of the implementation and almost everyone claiming any sort of real world success with coding agents seems to have reached a similar conclusion.
If you care about maintainability and quality (and I include maintaining using LLM based tools) then you need to understand what it does (in doing so you will find lots of things for it to fix - you'll probably find that the architecture it's chosen is not right for what you want too).
Tell a coding agent what your new thing needs to do, give it the absolute constraints, max response times, max failover times, and so on, tell it which technologies it has access to or could use, and then tell it to spend a lot of time going over and over the design, coming up with an initial X number of designs (I use 5), and then it must self criticise each one of them and weigh them up, narrow down to three, before finally presenting those three options to the user.
Now you read the options, understand them, realise that the AI has either converged on something very sensible, or it has missed something, so you tell it what it missed and iterate. Or it nailed something good, you pick the option you prefer, and tell it to come up with a more fleshed out high level design, describing the flow and behaviour deeply (NO CODE REFERENCES!). Then once you're happy, tell it to use that and write a comprehensive coding plan. Tell it specifically what coding patterns you prefer (you should have these in your AGENTS.md file already), what patterns to avoid (single threaded? multi-threaded? Avoid gc? How you typically deal with error conditions, etc etc).
Then have it start iteratively working on the coding plan, and it *MUST* have a strong feedback loop. If there is no feedback loop initially, I tell it to build one. It must be able to write very fluent integration tests (not just unit tests). It must be able to run the app and read the logs.
Do all this and I bet you get a better result that 80% of developers out there. Coding agents are extremely good when used well.
LLMs evaporated 90% of the "moments of despair" when you have an error and googling it isn't helping, or googling it made you realize you have to read 30min of documentation.
Coding is a joy now. LLMs shaved off all the rough edges.
I am terrified of allowing these things to complete tasks end-to-end with nothing intervening. Maybe that's why I don't run into many of these issues. I mostly delegate grunt work and manual tedium, not reasoning or design choices to the LLM. I may consult the LLM and ask for criticism, but there is no way I'm going to allow it to quietly make design decisions that I don't know about.
I use LLMs in the following ways:
1. Copy-pasting code into the web chat UI and asking for something (bugfix, add a feature, refactor, explain, review it etc), including entire source code files. A $20/mo Gemini subscription goes a long way (never been rate-limited). I only use the highest model. I often just copy-paste the entire source file between 3 backticks.
2. Cursor Tab. I do have hotkeys to enable and disable it; it's disabled most of the time otherwise it gets annoying.
3. Single-file changes directly from Cursor's AI sidebar. I only do this for simple, predictable stuff because even their auto-routing "Premium" setting is not as good as pasting stuff into Gemini 3.1 Pro.
That means I have only two $20/mo subscriptions: Gemini and Cursor.
I don't use Claude Code, it's really for people who don't know how to code. I don't use Plan Mode; I make and track the plan myself (if at all). I only tell the LLM granular tasks to execute. I don't use `claude.md` or `agents.md` or anything like that. If I don't like a particular output, I reset everything, modify my prompt and try again.
I believe this is the only way to fully leverage LLMs without losing any product quality. If you're trading off quality for "speed" (in quotes because over the long term, a low quality codebase is a massive drag on productivity) then there's no point.
Is that accurate?
What counts as “works” is the important bit, I think.
The LLM will do what you tell it to do. Manage it.
And then condensed an equal quantity of despair out of the ether via confident confabulations.
It's getting hard to keep up with trying to teach new devs what bad code looks like. And I swear sometimes they just copy my PR comments into their AI tool to fix the mistakes without any of the learning.
How have you set yours up that works well for you?
Take the bad result that you're getting, and pretend it's coming from an enthusiastic junior. What would you tell them to make them do this task better? Add that explanation to the agent (or explain that to the LLM and get it to add that to the agent, I have found this to work as well).
When you create a task for the LLM, get it to create a requirements document that lists all the requirements. Feed that into the review agent so it understands what the code agent was trying to do.
The LLM will do what you tell it to do. It doesn't magically understand what you want it to do. You have to tell it what to do.
A year ago I would've told my boss “can't be done” about my work today. I'd tell him to get me the right person to talk to (our partner, not an alien) who could give me some insight into what the hell I'm supposed to be doing to consume their API. Or to at least explain why it is that this can't be done.
Nowadays, I spent a couple of weeks reverse engineering their terrible ideas. Yeah, it worked. But it's a complete waste of my time, and tokens, energy, chips and RAM. And worst of all, it will lead to a terrible design.
That will work, but will eventually colapse under its own weight, as we use our increased power to increase our sloppiness and take it a little further. Because we can manage it. For now.
It writes the code for you. Then it runs the tests. Then it runs the linter. Then it runs the static analysis tool. If any of those fail, then it rewrites the code and runs them all again.
You only look at the code once it has done all of that.
If AI is ignoring the rest of the document and doing whatever it wants then you need to improve your document-writing skills. You can ask it why it did something, that helps discover how to improve. It's a process of refinement and discovery, just like learning how to use any new tool.
I use AI in two ways. In the way you describe, and also in the text editor as AI autocomplete. It works great until it doesn't. It inserts typos all the fucking time.
Also, don't assume you know it all.
I don't miss wasting an hour on a problem in a technology I'm not familiar with, where it's not like a big conceptual thing but something I could clear up in 5 seconds if I just had an expert in the room.
These LLMs can already incorporate our entire cultural corpus yet your "professional experience" is the threshold they won't cross?
Regarding such formal reasoning we have already seen marked improvement in the last year or two alone. The question is how this weighs on your prediction re their capabilities in the next two, five, ten, etc years.
The present notions of harnesses, structured output or looping in the LLM to some external state or sandbox be it debugger output or embedding into a runtime already show early promising results along these lines. I see no reason to believe these gains will not continue over the next five years.
If you have some theories in the converse in that regard I am all ears.
If you think the potential of LLMs is overblown feel free to short the market. I don't pretend to know the future. But if I may, I don't think you are framing the debate in the correct terms. Evidence is an important facet of human affairs. So is risk. Best of luck with your predictions.
I agree with you both - undoubtedly there are still massive gains to be made with the frontier models we have today with tooling and iteration, yet I do not believe there's sufficient evidence to claim we are rolling towards AG/SI on an exponential curve, without some additional breakthroughs given the jagged edges and data used to train models being fundamentally linear
You just need AI to be just good enough to win the tradeoff over a human employee. Just take your average office. Then ask yourself if the bar is really that high. AGI strikes me as an extremely nebulous concept. Better to just list everyone at your office and bucket them with a guess of how soon you think AI will replace them. Or weaken their market power. This is what every corporate boss in America is already doing. I'm merely suggesting rather than hope a graph curves in our individual favor we try to act more collectively as a species. Of course, I don't hold my breath.
I also don't find myself compelled by the notion that the danger to humanity is "AGI". The true danger is as it always has been - each other.
How many years away do you think we are from a “concierge” AI that can do the menial tasks handled by most personal assistants / program managers? Booking flights and hotels and coordinating employee availability?
“Unlimited progress” is not a statement on the rate of progress, it’s a statement on the limits of progress. It’s a much weaker claim than you’re framing it as. Your claim very much is that we have not yet reached the limits of LLMs potential. My claim, conversely, is that we’re already reaching diminishing returns, which are being masked by a massive influx of compute and energy. My short: LLMs are not the path to AGI.
Most likely because you haven't constrained their behavior in your prompt. You're making the assumption that they "understand" that using best practices is what you want. You have to tell them that, and tell them which practices they should use.
If incorrect LLM output is a prompt issue then demand for experienced developers will remain, and demand may actually increase as time passes.
If LLMs get good enough, one might be tempted to ask so what if most humans can't understand the output? Human civilization has by and large been a constant exercise in us collectively accomplishing more and more while individually comprehending less and less.
Our ancestors likely understood more about hunting live game or murdering each other than we do. Most of us do not consider that a great loss. Most of us living in the modern world depend on things we don't fully comprehend. I'm just not sure how this would lead to being reassured re the human as SWE.
The article you shared has little to do with this. Questions of how to divide up gains technology creates are a separate question from that of the technology itself. Tbh I found what you shared so boring I could barely finish it. I already in this thread made an exhortation to support politicans who commit to erasing inequality. The idea that LLMs can only exist with inequality is nonsensical. The only thing grim about what you shared is the lack of political imagination. It's boring.
Software specialization might look very different in 10 years but I doubt that technically specialized humans will be completely removed from their professions. We might not be carrying bows and arrows anymore but we will be carrying the equivalent of a rope and a Stetson.
I appreciate your points. I agree with you that not all "technically specialized humans will be completely removed" but let's not pretend the comparison is going from a caveman with a spear to a cowboy with a lasso. If you concede it is likely to be very different at some point calling it SWE is no longer useful.
I think SWEs would be better off realizing they have enjoyed a relatively extreme level of privilege, and rather than trying to hold onto it, use what time they still have to advocate for a more egalitarian society, even if that means giving up some of their gains. Otherwise speaking of farming, the mass layoffs to come when software has been disrupting blue collar jobs for decades will really be a chickens coming home to roost moment.
Considering that only 1% of the US workforce was a software engineer I expect similar workforce optimization to occur in software engineering specializations over the next 12,000 years. /s But seriously, it's never going to zero.
No need for specialized commercial software, if everyone can just explain to the computer what they want in English.
When working as a SWE, the longer I did it (~30 years) more of my time was spent understanding the problem, the edge cases, how to handle the edge cases, how to do all of it affordable, on time, and within budget.
That's engineering.
What you're describing is "writing code". That's lower value than "solving the problem".
I imagine a response, "But agile development, etc."
Yep. Part of solving the often sometimes involves creating prototypes to determine the essential viability of the solution. But that's only part of it. Which prototypes do you write? How much time do you allocate to same before accepting it's a dead end (at least for now) and punting on it?
That's engineering.
Me probably coming across as a dick today? Well, I was diagnosed autistic a year ago, and I'm on extended sabbatical/unemployment (3 years now) due to autistic burnout. And masking is part of how I got the burnout.**
* Why would someone be paying for that when there is likely someone else already doing it? Unless you're the rare person who hopes to "disrupt" the competition).
** has me begging the question of why I write here at all. SMH. Why do I do what I do? No idea sometimes.
There's the saying "Any idiot can build a bridge; it takes an engineer to build a bridge that barely stands."
To put this another way, any idiotic LLM can write code. It takes a person with domain experience to understand what code to write, rewrite, or not write.
I've seen lots of organizations hollow out their internal competence in favor of outsourcing the skills. LLMs are the ultimate expression of that. There are people who say "you need to have people in your organization who understand how things work because they're the ones who solve problems!" and there are other people who say "focus on your core competencies! These problems you're worried about aren't your core competencies, so get rid of those experts, they're expensive and annoying; we can just sign a contract with an organization that'll know things for us."
At some point we all will identify exactly how much "seed corn" you need for the next season. We'll figure that out because we're starving, but at least we'll all know.
got an email address in my profile if you'd be interested in talking at some point about something, or even talking about nothing in particular. (i don't normally do this sort of HN networking stuff, i find it super cringe. but there we go).
Or are you saying that I'm lying. That I am secretly hammering away at my keyboard while pretending not to?
No, writing code hasn't been how I spend most of my time for many decades now.
Write proper API documentation laying out the assumptions and intent, generate some good API docs, write a design and architecture document (which people find they need for LLMs to work at all anyway). You’ll find that you spend a lot less time reading code.
Everything we have to do for AI to function well, would help humans to function better too.
If you take the things for AI, but do them for humans instead, that human will easily 2x or more, and someone will actually understand the code that gets written.
This only works on high-trust teams and organizations. A lot of AI productivity gains are from SWE putting the extra effort because the results will be attributed to them. Being a force-multiplier for others isn't always recognized, instead, your perfomance will likely judged solely on the metrics directly attributed to you. I learned this lesson the hard way by being idealistic, and overestimating the level of trust that had been built after joining a new team. Companies pay lip service to software quality, no one gives a shit if your code has the lowest SEV rates.
Writing code just isn't what takes time.
This is how big corporations look like, not some SV startups.
(*) Well, now claude spends a lot of time writing code, I spend a lot of time designing and steering it. Claude can write remarkably sophisticated code with the correct steering.*
Less capable engineers think in terms of implementing runtime execution and solving runtime errors.
More capable engineers think in terms of designing the most effective architecture and abstractions for maintainability, performance, and robustness.
I have a project I am working on, It has not compiled in months, but that's ok, since the real work, for me, is in the architectural design.
Yes, getting it to actually run takes some time and effort, but for me that is almost mechanical now.
Maybe you will still be needed. That is one question. How well you will be paid and treated when the barrier to entry is now "I can think" is another. As the parent indicates, most people doing software are not doing things akin to pure math. I don't think most SWEs want that lifestyle anyway.
It's ok. You shouldn't fight the coming change. Instead use the time we still have to fight for more equal outcomes (vote for politicians that support UBI, Medicare for all). The longer you delude yourself that you are uniquely needed in an increasingly mechanized world the worse all our outcomes will be.
The progress models have made in the last 5 years aren't convincing me they'll bridge that gap too soon, although I can see how some people are convinced by how decent agentic harnesses make things. I know it's really easy to get very hyped with the current state of the technology, but try to have a bit of skepticism.
Those two formulations represent different developers' approaches to the same task. The former being developers who are much better at planning than the latter.
> Most of the time is spent coding which encompasses typing, retyping, and retyping again. It also includes banging your head against the wall while trying to get one of your rewrites to work against and under-documented API.
I don't think I've experienced this to a large degree. Maybe early in my career. Most of my time now is spent formulating a solution, and time spent coding is mostly spent trying to compose my changes with the existing code in a way that is performant, reliable and meets the specifications.
One wonders why AI hasn't replaced all those non-existent documentation writers yet.
Therein lies a clue to what the future holds.
The actual problem solving was trivial but I would spend days trying to work my way around some Qt work or guess the magic Cmake incantation.
You tried to make it sound like a far more noble and risky thing than it is.
OP's formulation makes SWE sound like a purely noble enterprise like mathematics. It's more like an oil rig worker banging on pieces of metal with large hammers to get the drill string put together. They went in with a plan, but the reality didn't agree and they are on a tight schedule.