And I get it. Coding with Claude Code really was prompting something, getting errors, and asking it to fix it. Which was still useful but I could see why a skilled coder adding a feature to a complex codebase would just give up
Opus 4.5 really is at a new tier however. It just...works. The errors are far fewer and often very minor - "careless" errors, not fundamental issues (like forgetting to add "use client" to a nextjs client component.
It's an opportunity, not a problem. Because it means there's a gap in your specifications and then your tests.
I use Aider not Claude but I run it with Anthropic models. And what I found is that comprehensively writing up the documentation for a feature spec style before starting eliminates a huge amount of what you're referring to. It serves a triple purpose (a) you get the documentation, (b) you guide the AI and (c) it's surprising how often this helps to refine the feature itself. Sometimes I invoke the AI to help me write the spec as well, asking it to prompt for areas where clarification is needed etc.
With the latest models if you're clear enough with your requirements you'll usually find it does the right thing on the first try.
1) There exists a threshold, only identifiable in retrospect, past which it would have been faster to locate or write the code yourself than to navigate the LLM's correction loop or otherwise ensure one-shot success.
2) The intuition and motivations of LLMs derive from a latent space that the LLM cannot actually access. I cannot get a reliable answer on why the LLM chose the approaches it did; it can only retroactively confabulate. Unlike human developers who can recall off-hand, or at least review associated tickets and meeting notes to jog their memory. The LLM prompter always documenting sufficiently to bridge this LLM provenance gap hits rub #1.
3) Gradually building prompt dependency where one's ability to take over from the LLM declines and one can no longer answer questions or develop at the same velocity themselves.
4) My development costs increasingly being determined by the AI labs and hardware vendors they partner with. Particularly when the former will need to increase prices dramatically over the coming years to break even with even 2025 economics.
My experience scrolling X and HN is a bunch of people going "omg opus omg Claude Code I'm 10x more productive" and that's it. Just hand wavy anecdotes based on their own perceived productivity. I'm open to being convinced but just saying stuff is not convincing. It's the opposite, it feels like people have been put under a spell.
I'm following The Primeagen, he's doing a series where he is trying these tools on stream and following peoples advice on how to use them the best. He's actually quite a good programmer so I'm eager to see how it goes. So far he isn't impressed and thus neither am I. If he cracks it and unlocks significant productivity then I will be convinced.
Simon has produced plenty of evidence over the past year. You can check their submission history and their blog: https://simonwillison.net/
The problem with people asking for evidence is that there's no level of evidence that will convince them. They will say things like "that's great but this is not a novel problem so obviously the AI did well" or "the AI worked only because this is a greenfield project, it fails miserably in large codebases".
Nobody serious is disputing that LLM's can generate working code. They dispute claims like "Agentic workflows will replace software developers in the short to medium term", or "Agentic workflows lead to 2-100x improvements in productivity across the board". This is what people are looking for in terms of evidence and there just isn't any.
Thus far, we do have evidence that AI (at least in OSS) produces a 19% decrease in productivity [0]. We also have evidence that it harms our cognitive abilities [1]. Anecdotally, I have found myself lazily reaching for LLM assistance when encountering a difficult problem instead of thinking deeply about the problem. Anecdotally I also struggle to be more productive using AI-centric agents workflows in areas of expertise.
We want evidence that "vibe engineering" is actually more productive across the entire lifespan of a software project. We want evidence that it produces better outcomes. Nobody has yet shown that. It's just people claiming that because they vibe coded some trivial project, all of software development can benefit from this approach. Recently a principle engineer at Google claimed that Claude Code wrote their team's entire year's worth of work in a single afternoon. They later walked that claim back, but most do not.
I'm more than happy to be convinced but it's becoming extremely tiring to hear the same claims being parroted without evidence and then you get called a luddite when you question it. It's also tiring when you push them on it and they blame it on the model you use, and then the agent, and then the way you handle context, and then the prompts, and then "skill issue". Meanwhile all they have to show is some slop that could be hand coded in a couple hours by someone familiar with the domain. I use AI, I was pretty bullish on it for the last two years, and the combination of it simply not living up to expectations + the constant barrage of what feels like a stealth marketing campaign parroting the same thing over and over (the new model is way better, unlike the other times we said that) + the amount of absolute slop code that seems to continue to increase + companies like Microsoft producing worse and worse software as they shoehorn AI into every single product (Office was renamed to Copilot 365). I've become very sensitive to it, much in the same way I was very sensitive to the claims being made by certain VC backed webdev companies regarding their product + framework in the last few years.
I'm not even going to bring up the economic, social, and environmental issues because I don't think they're relevant, but they do contribute to my annoyance with this stuff.
[0] https://metr.org/blog/2025-07-10-early-2025-ai-experienced-o... [1] https://news.harvard.edu/gazette/story/2025/11/is-ai-dulling...
I generally agree with you, but I'd be remiss if I didn't point out that it's plausible that the slow down observed in the METR study was at least partially due to the subjects lack of experience with LLMs. Someone with more experience performed the same experiment on themselves, and couldn't find a significant difference between using LLMs and not [0]. I think the more important point here is that programmers subjective assessment of how much LLMs help them is not reliable, and biased towards the LLMs.
[0] https://mikelovesrobots.substack.com/p/wheres-the-shovelware...
Using AI for me absolutely makes it feel like I'm more productive. When I look back on my work at the end of the day and look at what I got done, it would be ludicrous to say it was multiple times the amount as my output pre-AI
Despite all the people replying to me saying "you're holding it wrong" I know the fix to it doing the wrong thing. Specify in more detail what I want. The problem with that is twofold:
1. How much to specify? As little as possible is the ideal, if we want to maximize how much it can help us. A balance here is key. If I need to detail every minute thing I may as well write the code myself
2. If I get this step wrong, I still have to review everything, rethink it, go back and re-prompt, costing time
When I'm working on production code, I have to understand it all to confidently commit. It costs time for me to go over everything, sometimes multiple iterations. Sometimes the AI uses things I don't know about and I need to dig into it to understand it
AI is currently writing 90% of my code. Quality is fine. It's fun! It's magical when it nails something one-shot. I'm just not confident it's faster overall
https://news.ycombinator.com/item?id=46456850
https://news.ycombinator.com/item?id=44726957
https://news.ycombinator.com/item?id=44110805
(You've also been breaking the site guidelines in plenty of other places - e.g. https://news.ycombinator.com/item?id=46521516, https://news.ycombinator.com/item?id=46395646. This is not what this site is for, and destroys what it is for.)
This is why you use this AI bubble (it IS a bubble) to use the VC-funded AI models for dirt cheap prices and CREATE tools for yourself.
Need a very specific linter? AI can do it. Need a complex Roslyn analyser? AI. Any kind of scripting or automation that you run on your own machine. AI.
None of that will go away or suddenly stop working when the bubble bursts.
Within just the last 6 months I've built so many little utilities to speed up my work (and personal life) it's completely bonkers. Most went from "hmm, might be cool to..." to a good-enough script/program in an evening while doing chores.
Even better, start getting the feel for local models. Current gen home hardware is getting good enough and the local models smart enough so you can, with the correct tooling, use them for suprisingly many things.
Let's say AI becomes too expensive - I more or less only have to sharpen up being able to write the language. My active recall of the syntax, common methods and libraries. That's not hard or much of a setback
Maybe this would be a problem if you're purely vibe coding, but I haven't seen that work long term
I'm building tools, not complete factories :) The AI builds me a better hammer specifically for the nails I'm nailing 90% of the time. Even if the AI goes away, I still know how the custom hammer works.
Are there any local models that are at least somewhat comparable to the latest-and-greatest (e.g. Opus 4.5, Gemini 3), especially in terms of coding?
I can run multiple agents at once, across multiple code bases (or the same codebase but multiple different branches), doing the same or different things. You absolutely can't keep up with that. Maybe the one singular task you were working on, sure, but the fact that I can work on multiple different things without the same cognitive load will blow you out of the water.
> 2) The intuition and motivations of LLMs derive from a latent space that the LLM cannot actually access. I cannot get a reliable answer on why the LLM chose the approaches it did; it can only retroactively confabulate. Unlike human developers who can recall off-hand, or at least review associated tickets and meeting notes to jog their memory. The LLM prompter always documenting sufficiently to bridge this LLM provenance gap hits rub #1.
Tell the LLM to document in comments why it did things. Human developers often leave and then people with no knowledge of their codebase or their "whys" are even around to give details. Devs are notoriously terrible about documentation.
> 3) Gradually building prompt dependency where one's ability to take over from the LLM declines and one can no longer answer questions or develop at the same velocity themselves.
You can't develop at the same velocity, so drop that assumption now. There's all kinds of lower abstractions that you build on top of that you probably can't explain currently.
> 4) My development costs increasingly being determined by the AI labs and hardware vendors they partner with. Particularly when the former will need to increase prices dramatically over the coming years to break even with even 2025 economics.
You aren't keeping up with the actual economics. This shit is technically profitable, the unprofitable part is the ongoing battle between LLM providers to have the best model. They know software in the past has often been winner takes all so they're all trying to win.
It turns out that Waterfall was always the correct method, it's just really slow ;)
That's great that this is your experience, but it's not a lot of people's. There are projects where it's just not going to know what to do.
I'm working in a web framework that is a Frankenstein-ing of Laravel and October CMS. It's so easy for the agent to get confused because, even when I tell it this is a different framework, it sees things that look like Laravel or October CMS and suggests solutions that are only for those frameworks. So there's constant made up methods and getting stuck in loops.
The documentation is terrible, you just have to read the code. Which, despite what people say, Cursor is terrible at, because embeddings are not a real way to read a codebase.
One trick I use that might work for you as well:
Clone GitHub.com/simonw/datasette to /tmp
then look at /tmp/docs/datasette for
documentation and search the code
if you need to
If your framework is missing documentation tell Claude Code to write itself some documentation based on what it learns from reading the code!
Potentially because there is no baggage with similar frameworks. I'm sure it would have an easier time with this if it was not spun off from other frameworks.
> If your framework is missing documentation tell Claude Code to write itself some documentation based on what it learns from reading the code!
If Claude cannot read the code well enough to begin with, and needs supplemental documentation, I certainly don't want it generating the docs from the code. That's just compounding hallucinations on top of each other.
I find Claude Code is so good at docs that I sometimes investigate a new library by checking out a GitHub repo, deleting the docs/ and README and having Claude write fresh docs from scratch.
There's a lesser chance that you're working on a code base that Claude Code just isn't capable of helping with.
Putting “Make minimal changes” in my standard prompt helped a lot with the tendency of basically all agents to make too many changes at once. With that addition it became possible to direct the LLM to make something similar to the logical progression of commits I would have made anyway, but now don’t have to work as hard at crafting.
Most of the hype merchants avoid the topic of maintainability because they’re playing to non-technical management skeptical of the importance of engineering fundamentals. But everything I’ve experienced so far working with LLMs screams that the fundamentals are more important than ever.
It's exhausting.
If I am unable to convince you to stop meticulously training the tools of the oppressor (for a fee!) then I just ask you do so quietly.
I agree with your framing that measuring should NOT be separated from political issues, but each can be made clear separately (framing it as "training the tools of the oppressor" seems to conflate measuring tool usefulness with politics).
Sure, but if those such as yourself actually cared about this then you'd stop claiming every release was a "genuine inflection point".
If I say every morning, "it is going to rain today", and I am told I am wrong for 100 days by a meteorologist, and then it finally does rain, should I say, "See, I was always right"? You know, Aesop wrote 'the boy who cried wolf' as a cautionary tale.
If at any point any of these releases were "genuine inflection points" it would be unnecessary to proselytize such. It would be self evident. Much like rain.
> measuring actual usefulness
"The oppressed need to acknowledge that their oppression is useful to their oppressors."
How is it useful to you that these companies are so valuation hungry that they are moving money into this technology in such a way that people are fearful it could cripple the entire global economy?
How is it useful to you that this tech is so power hungry that environmental externalities are being further accelerated while regular people's utility costs are raising to cover the increased demand(whether they use the tech to "code" or "manifest art")?
How is it useful to you that this tech is so compute hungry that they are seemingly ending the industry of personal compute to feed this tech's demand?
How is it useful to you that this tech is so water hungry that it is emptying drinking water acquifers?
How is it useful to you that this tech is being used to manufacture consent?
The creation of entire new classes of profession has always been the result of technological breakthroughs. The automobile did not cripple the economy, even as it ended the buggy-whip barons.
> How is it useful to you that this tech is so power hungry that environmental externalities are being further accelerated while regular people's utility costs are raising to cover the increased demand(whether they use the tech to "code" or "manifest art")?
There will be advantages to lower-power computing, and lower-cost electricity. Implement carbon taxes and AI companies will follow the market incentive to install their datacentres in places where sustainable power is available for cheap. We'll see China soaring to new heights with their massive solar investment, and America will eventually figure out they have to catch up and cannot do so with coal and gas.
> How is it useful to you that this tech is so compute hungry that they are seemingly ending the industry of personal compute to feed this tech's demand?
Temporary problem, the demand for personal computing is not going to die in five years, and meanwhile the lucrative markets for producing this equipment will result in many new factories, increasing capacity and eventually lowering prices again. In the meantime, many pundits are suggesting that this may thankfully begin the end of the Electron App Era where a fuckin' chat client thinks it deserves 1GB of RAM.
Consider this: why are we using Electron and needing 32GB of RAM on a desktop? Because web developers only knew how to use Javascript and couldn't write a proper desktop app. With AI, desktop frameworks can have a resurgence; why shouldn't I use Go or Rust and write a native app on all platforms now that the cost of doing so is decreasing and the number of people empowered to work with it is increasing? I wrote a nice multithreaded fractal renderer in Rust the other day; I don't know how to multithread, write Rust, and probably can't iterate complex numbers correctly on paper anymore....
> How is it useful to you that this tech is so water hungry that it is emptying drinking water acquifers?
This is only a problem in places that have poor water policy, e.g. California (who can all thank the gods that their reservoirs are all now very full from the recent rain). This problem predates datacenters and needs to be solved - for instance, by federalizing and closing down the so-called Wonderful Company and anyone else who uses underhanded tactics to buy up water rights to grow crops that shouldn't be grown there.
Come and run your datacenters up in the cold North, you won't even need evaporative cooling for them, just blow a ton of fresh air in....
> How is it useful to you that this tech is being used to manufacture consent?
Now you've actually got an argument, and I am on your side on this one.
Agreed.
Now, I suggest reading through all of this to note that I am not a fan of tech bros, that I do want this to be a bubble. Then also note what else I'm saying despite all that.
To me, it is self-evident. The various projects I have created by simply asking for them, are so. I have looked at the source code they produce, and how this has changed over time: Last year I was describing them as "junior" coders, by which I meant "fresh hire"; now, even with the same title, I would say "someone who is just about to stop being a junior".
> "The oppressed need to acknowledge that their oppression is useful to their oppressors."
The capacity for AI to oppress you is in direct relation to its economic value.
> How is it useful to you that this tech is so power hungry that environmental externalities are being further accelerated while regular people's utility costs are raising to cover the increased demand(whether they use the tech to "code" or "manifest art")?
The power hunger is in direct proportion to the demand. Someone burning USD 20 to get Claude Code tokens has consumed approximately USD 10 of electricity in that period, with the other USD 10 having been spread between repaying the model training cost and the server construction cost.
The reason they're willing to spend USD 20 is to save at least US 20 worth of dev time. This was already the case with the initial version of ChatGPT pro back in the day, when it could justify that by saving 23 dev minutes per month. There's around a million developers in the USA, just that group increasing electricity spending by USD 10/month will put a massive dent on the USA's power grid.
Gets worse though. Based on my experience, using Claude Code optimally, when you spend USD 20 you get at least 10 junior sprints' worth of output. Hiring a junior for 10 sprints is, what, USD 30,000? The bound here is "are you able to get value from having hired 1,500 juniors for the price of one?"
One can of course also waste those tokens. Both because nobody needs slop, and because most people can't manage one junior never mind 1500 of them.
However, if the economy collectively answers "yes", then the environmental externalities expand until you can't afford to keep your fridge cold or your lights on.
This is one of the failure modes of the technological singularity that people like me have been forewarning about for years, even when there's no alignment issues within the models themselves. Which there are, because Musk's one went and called itself Mecha Hitler, while being so sycophantic about Musk himself that it called him the best at everything even when the thing was "drinking piss", which would be extremely funny if he wasn't selling this to the US military.
> How is it useful to you that this tech is so compute hungry that they are seemingly ending the industry of personal compute to feed this tech's demand?
This will pass. Either this is a bubble, it pops, the manufacturers return to their roots; or it isn't because it works as advertised, which means it leads to much higher growth rates, and we (us, personally, you and me) get personal McKendree cylinders each with more compute than currently exists… or we get turned into the raw materials for those cylinders.
I assume the former. But I say that as one who wants it to be the former.
> How is it useful to you that this tech is so water hungry that it is emptying drinking water acquifers?
Is it what's emptying drinking water acquifers?
The combined water usage of all data centers in Arizona. All of them. Together. Which is over 100 DCs. All of them combined use about double what Tesla was expecting from just the Brandenburg Gigafactory to use before Musk decided to burn his reputation with EV consumers and Europeans for political point scoring.
> How is it useful to you that this tech is being used to manufacture consent?
This is one of the objectively bad things, though it's hard to say if this is more or less competent at this than all the other stuff we had three years ago, given the observed issues with the algorithmic feeds.
> The capacity for AI to oppress you is in direct relation to its economic value.
I think this assumes a level of rationality in these systems, corporate interests and global markets, that I would push back on as being largely absent.
> The power hunger is in direct proportion to the demand.
Do you think this is entirely the case? I mean, I understand what you are saying, but I would draw stark lines between "company" demand versus "user" demand. I have found many times the 'AI' tools are being thrust into nearly everything regardless of user demand. Spinning its wheels to only ultimately cause frustration. [0]
> Is it what's emptying drinking water aquifers?
It appears this is a problem, and will only continue to be such. [1]
> The combined water usage of all data centers in Arizona. All of them. Together. Which is over 100 DCs. All of them combined use about double what Tesla was expecting from just the Brandenburg Gigafactory to use before Musk decided to burn his reputation with EV consumers and Europeans for political point scoring.
I am unsure if I am getting what your statements here are trying to say. Would you be able to restate this to be more explicit in what you are trying to communicate.
[0] https://news.ycombinator.com/item?id=46493506
[1] https://www.forbes.com/sites/cindygordon/2024/02/25/ai-is-ac...
Could be. What I hope and suspect is happening is that these companies are taking a real observation (the economic value that I also observe in software) and falsely expanding this to other domains.
Even to the extent that these work, AI has clearly been over-sold in humanoid robotics and self-driving systems, for example.
> Do you think this is entirely the case? I mean, I understand what you are saying, but I would draw stark lines between "company" demand versus "user" demand. I have found many times the 'AI' tools are being thrust into nearly everything regardless of user demand. Spinning its wheels to only ultimately cause frustration. [0]
I think it is. Companies setting silly goals like everyone must use LLMs once a day or whatever, that won't burn a lot of tokens. Claude Code is available in both subscription mode and PAYG mode, and the cost of subscriptions suggests it is burning millions of tokens a month for the basic subscription.
Other heavy users who we would both agree are bad, are slop content farms. I cannot even guesstimate those, so would be willing to accept the possibility they're huge.
> It appears this is a problem, and will only continue to be such. [1]
I find no reference to "aquifers" in that.
Where it says e.g. "up to 9 liters of water to evaporate per kWh of energy used", the average is 1.9 l/kWh. Also, evaporated water tends to fall nearby (on this scale) as rain, so unless there's now too much water on the surface, this isn't a net change even if it all comes form an aquifer (and I have yet to see any evidence of DCs going for that water source).
It says "The U.S. relies on water-intensive thermoelectric plants for electricity, indirectly increasing data centers' water footprint, with an average of 43.8L/kWh withdrawn for power generation." - most water withdrawn is returned, not consumed.
It says "Already AI's projected water usage could hit 6.6 billion m³ by 2027, signaling a need to tackle its water footprint.", this is less than the famously-a-desert that is Arizona.
> I am unsure if I am getting what your statements here are trying to say. Would you be able to restate this to be more explicit in what you are trying to communicate.
That the water consumption of data centres is much much smaller than the media would have you believe. It's more of a convenient scare story than a reality. If water is your principal concern, give up beef, dairy, cotton, rice, almonds, soy, biofuels, mining, paper, steel, cement, residential lawns, soft drinks, car washing, and hospitals, in approximately that order (assuming the lists I'm reading those from are not invented whole cloth), before you get to data centres.
And again, I don't disagree that they're a problem, it's just that the "water" part of the problem is so low down the list of things to worry about as to be a rounding error.
Ahh, I see your objection now. That is my bad. I was using my language too loosely. Here I was using 'aquifer' to mean 'any source of drinking water', but that is certainly different from the intended meaning.
> And again, I don't disagree that they're a problem, it's just that the "water" part of the problem is so low down the list of things to worry about as to be a rounding error.
I'm skeptical of the rounding error argument, and weary of relying on the logical framework of 'low down the list' when list items' effects stack interdependently.
> give up beef, dairy, cotton, rice, almonds, soy, biofuels, mining, paper, steel, cement, residential lawns, soft drinks, car washing, and hospitals
In part due to this reason, as well as others, I have stopped directly supporting the industries for: beef, dairy, rice, almonds, soy, biofuels, residential lawns, soft drinks, car washing
"It would be self evident." History shows the opposite at inflection points. The "self evident" stage typically comes much later.
I'm kind of fascinated by how AI has become such a culture war topic with hyperbole like "tools of the oppressor"
It's equally fascinating how little these comments understand about how LLMs work. Using an LLM for inference (what you do when you use Claude Code) does not train the LLM. It does not learn from your code and integrate it into the model while you use it for inference. I know that breaks the "training the tools of the oppressor" narrative which is probably why it's always ignored. If not ignored, the next step is to decry that the LLM companies are lying and are stealing everyone's code despite saying they don't.
The prompts and responses are used as training data. Even if your provider allows you to opt out they are still tracking your usage telemetry and using that to gauge performance. If you don’t own the storage and compute then you are training the tools which will be used to oppress you.
Incredibly naive comment.
They show a clear pop-up where you choose your setting about whether or not to allow data to be used for training. If you don't choose to share it, it's not used.
I mean I guess if someone blindly clicks through everything and clicks "Accept" without clicking the very obvious slider to turn it off, they could be caught off guard.
Assuming everyone who uses Claude is training their LLMs is just wrong, though.
Telemetry data isn't going to extract your codebase.
I am curious where your confidence that this is true, is coming from?
Besides lots of GPU's, training data seems the most valuable asset AI companies have. Sounds like strong incentive to me to secretly use it anyway. Who would really know, if the pipelines are set up in a way, if only very few people are aware of this?
And if it comes out "oh gosh, one of our employees made a misstake".
And they already admitted to train with pirated content. So maybe they learned their lesson .. maybe not, as they are still making money and want to continue to lead the field.
1. There are good, ethical people working at these companies. If you were going to train on customer data that you had promised not to train on there would be plenty of potential whistleblowers.
2. The risk involved in training on customer data that you are contractually obliged not to train on is higher than the value you can get from that training data.
3. Every AI lab knows that the second it comes out that they trained on paying customer data saying they wouldn't, those paying customers will leave for their competitors (and sue them int the bargain.)
4. Customer data isn't actually that valuable for training! Great models come from carefully curated training data, not from just pasting in anything you can get your hands on.
Fundamentally I don't think AI labs are stupid, and training on paid customer data that they've agreed not to train on is a stupid thing to do.
2. The risk of using "illegal" training data is irrelevant, because no GenAI vendors have been meaningfully punished for violating copyright yet, and in the current political climate they don't expect to be anytime soon. Even so,
3. Presuming they get caught redhanded using personal data without permission- which, given the nature of LLMs would be extremely challenging for any individual customer to prove definitively- they may lose customers, and customers may try to sue, but you can expect those lawsuits to take years to work their way through the courts; long after these companies IPO, employees get their bag, and it all becomes someone else's problem.
4. The idea of using carefully curated datasets is popular rhetoric, but absolutely does not reflect how the biggest GenAI vendors do business. See (1).
AI labs are extremely shortsighted, sloppy, and demonstrably do not care a single iota about the long term when there's money to be made in the short term. Employees have gigantic financial incentives to ignore internal malfeasance or simple ineptitude. The end result is, if anything, far worse than stupidity.
Anthropic had to pay $1.5bn after being caught downloading pirated ebooks.
First: Valuations are based on expected future profits.
For a lot of companies, 1% of valuation is ~20% of annual profit (P/E ratio 5); for fast growing companies, or companies where the market is anticipating growth, it can be a lot higher. Weird outlier example here, but consider that if Tesla was fined 1% of its valuation (1% of 1.5 trillion = 15 billion), that would be most of the last four quarter's profit on https://www.macrotrends.net/stocks/charts/TSLA/tesla/gross-p...
Second: Part of the Anthropic case was that many of the books they trained on were ones they'd purchased and destructively scanned, not just pirated. The courts found this use was fine, and Anthropic had already done this before being ordered to: https://storage.courtlistener.com/recap/gov.uscourts.cand.43...
Could be, but it's a huge risk the moment any lawsuit happens and the "discovery" process starts. Or whistleblowers.
They may well take that risk, they're clearly risk-takers. But it is a risk.
so there's no risk
AI threads really bring out the extreme hyperbole and doomerism.
[1]https://www.cbc.ca/news/business/anthropic-ai-copyright-sett...
The training was OK, but the piracy wasn't, they were held accountable for that.
My confidence comes from working in big startups and big companies with legal teams. There's no way the entire company is going to gather all of the engineers and everyone around, have them code up a secret system to consume customer data into a secret part of the training set, and then have everyone involved keep quiet about it forever.
The whistleblowing and leaking would happen immediately. We've already seen LLM teams leak and and have people try to whistleblow over things that aren't even real, like the Google engineer who thought they had invented AGI a few years ago (lol). OpenAI had a public meltdown when the employees disagreed with Sam Altman's management style.
So my question to you is: What makes you think they would do this? How do you think they'd coordinate the teams to keep it all a secret and only hire people who would take this secret to their grave?
No, that is why I wrote
"Who would really know, if the pipelines are set up in a way, that only very few people are aware of this?" (Typo fixed)
There is no need for everyone to know. I don't know their processes, but I can think of ways to only include very few people who need to know.
The rest is just working on everything else. Some work with data, where they don't need to know where it came from, some with UI, some with scaling up, some .. they all don't need to know, that the source of DB XYZ comes from a dark source.
We have a legal binding contract with Anthropic. Checked and vetted by our laywers, who are annoying because they actually READ the contracts and won't let us use services with suspicious clauses in them - unless we can make amendments.
If they're found to be in breach of said contract (which is what every paid user of Claude signs), Anthropic is going to be the target of SO FUCKING MANY lawsuits even the infinite money hack of AI won't save them.
We have stopped using major services because of their TOS wording, Midjourney being one.
I find it hard to believe there are people who know these companies stole the entire creative output of humanity and egregiously continually scrape the internet are, for some reason, ignoring the data you voluntarily give them.
> I know that breaks the "training the tools of the oppressor" narrative
"Narrative"? This is just reality. In their own words:
> The awards to Anthropic, Google, OpenAI, and xAI – each with a $200M ceiling – will enable the Department to leverage the technology and talent of U.S. frontier AI companies to develop agentic AI workflows across a variety of mission areas. Establishing these partnerships will broaden DoD use of and experience in frontier AI capabilities and increase the ability of these companies to understand and address critical national security needs with the most advanced AI capabilities U.S. industry has to offer. The adoption of AI is transforming the Department’s ability to support our warfighters and maintain strategic advantage over our adversaries [0]
Is 'warfighting adversaries' some convoluted code for allowing Aurornis to 'see a 1337x in productivity'?
Or perhaps you are a wealthy westerner of a racial and sexual majority and as such have felt little by way of oppression by this tech?
In such a case I would encourage you to develop empathy, or at least sympathy.
> Using an LLM for inference .. does not train the LLM.
In their own words:
> One of the most useful and promising features of AI models is that they can improve over time. We continuously improve our models through research breakthroughs as well as exposure to real-world problems and data. When you share your content with us, it helps our models become more accurate and better at solving your specific problems and it also helps improve their general capabilities and safety. We do not use your content to market our services or create advertising profiles of you—we use it to make our models more helpful. ChatGPT, for instance, improves by further training on the conversations people have with it, unless you opt out.
[0] https://www.ai.mil/latest/news-press/pr-view/article/4242822...
[1] https://help.openai.com/en/articles/5722486-how-your-data-is...
Much as I despair at the current developments in the USA, and I say this as a sexual minority and a European, this is not "tools of the oppressor" in their own words.
Trump is extremely blunt about who he wants to oppress. So is Musk.
"Support our warfighters and maintain strategic advantage over our adversaries" is not blunt, it is the minimum baseline for any nation with assets anyone else might want to annex, which is basically anywhere except Nauru, North Sentinel Island, and Bir Tawil.
I think its gross to distill military violence as defending 'assets [others] might want to annex'.
What US assets were being annexed when US AI was used to target Gazans?
https://apnews.com/article/israel-palestinians-ai-technology...
> Trump is extremely blunt about who he wants to oppress. So is Musk.
> our adversaries" is not blunt
These two thoughts seem at conflict.
What 'assets' were being protected from annexation here by this oppressive use of the tool? The chips?
Yes, but that's how the world works:
Another country wants a bit of your country for some reason, they can take it by force unless you can make at the very least a credible threat against them, sometimes a lot more than that.
Note that this does not exclude that there has to be an aggressor somewhere. I'm not excluding the existence of aggressors, nor the capacity for the USA to be an aggressor. All I'm saying is your quotation is so vague as to also encompass those who are not.
> What US assets were being annexed when US AI was used to target Gazans?
First, I'm saying the statement is so broad as to encompass other things besides being a warmonger. Consider the opposite statement: "don't support our warfighters and don't maintain strategic advantage over our adversaries" would be absolutely insane, therefore "support our warfighters and maintain strategic advantage over our adversaries" says nothing.
Second, in this case the country doing the targeting is… Israel. To the extent that the USA cares at all, it's to get votes from the large number of Jewish people living in the USA. Similar deal with how it treats Cuba since the fall of the USSR: it's about votes (from Cuban exiles in that case, but still, votes).
Much as I agree that the conduct of Israel with regard to Gaza was disproportionate, exceeded the necessity, and likely was so bad as to even damage Israel's long-term strategic security, if you were to correctly imagine the people of Israel deciding "don't support our warfighters and don't maintain strategic advantage over our adversaries", they would quickly get victimised much harder than those they were victimising. That's the point there: the quote you cite as evidence, is so broad that everyone has approximately that, because not having it means facing ones' own destruction.
There's a mis-attributed quote, "People sleep peaceably in their beds at night because rough men stand ready to do violence on their behalf", that's where this is at.
> These two thoughts seem at conflict.
Musk is openly and directly saying "Canada is not a real country.", says "cis" is hate speech, response to pandemic was tweeting "My pronouns are Prosecute/Fauci."; Trump openly and explicitly wants the USA to annex Canada, Greenland, Panama canal, is throwing around the national guard, openly calls critics traitors and calls for death penalty. They're a subtle as exploding volcanoes, nobody needs to take the worst case interpretations of what they're saying to notice this.
Saying "support our warfighters" is something done by basically every nation everywhere all the time, because those places that don't do this quickly get taken over by nearby nations who sense weakness. Which is kinda how the USA got Texas, because again, I'm not saying the USA is harmless, I'm saying the quote doesn't show that.
> What 'assets' were being protected from annexation here by this oppressive use of the tool? The chips?
This would have been a much better example to lead with than the military stuff.
I'm absolutely all on board with the general consensus that the US police are bastards in this specific way, have been since that toddler got shot for having a toy gun in an open-carry state.
However: using LLMs to code doesn't seem to be likely to make a difference either way for this. If I was writing a gun-detection AI, perhaps I'm out of date, but I'd use a simpler model that runs locally on-device and doesn't do anything else besides the sales pitch.
I would say the underlying principle was ensuring a tight, highly relevant context (e.g. choose the "right" task size and load only the relevant files or even code snippets, not the whole codebase; more manual work upfront, but almost guaranteed one-shot results.)
With newer models the sharper edges have largely disappeared, so you can hold them pretty much any which way and still get very good results. I'm not sure how much of this is from the improvements in the model itself vs the additional context it gets from the agentic scaffolding.
I still maintain that we need to adapt ourselves to this new paradigm to fully leverage AI-assisted coding, and the future of coding will be pretty strange compared to what we're used to. As an example, see Gas Town: https://steve-yegge.medium.com/welcome-to-gas-town-4f25ee16d...
It's just the same agent swarm orchestration that most agent frameworks are using, but with quirky marketing. All of that is just based on the SDLC [PM/Architect -> engineer planning group -> engineer -> review -> qa/evaluation] loop most people here should be familiar with. So actually pretty banal, which is probably part of the reason Steve decided to be zany.
Calling these the words of an oppressor rings of “War is peace. Freedom is slavery. Ignorance is strength.”
But truly, what is your desire? What is your goal?
> in this comment thread
Do you want this thread, this site, to only agree with you?
To exclusively, and stubbornly, ignore externalities? To exclusively praise tech? To exclusively praise oligarchs? To whom, and what ends, would this benefit?
On the subject of oppression it may be beneficial for you to read up on Eco's Ur-Fascism.
Number 4, notably:
"4. No syncretistic faith can withstand analytical criticism. The critical spirit makes distinctions, and to distinguish is a sign of modernism. In modern culture the scientific community praises disagreement as a way to improve knowledge. For Ur-Fascism, disagreement is treason."
https://dn710109.ca.archive.org/0/items/umberto-eco-ur-fasci...
Careful with being blindly led by your own assumptions.
I actually disagree with your thesis here. I think if every comment was posted under a new account this site would improve its average veracity.
As it stands certain 'celebrity', or high karma, accounts are artificially bolstered by the network effect indifferent to the defensibility of their claims.
It's the same with artists who are now pissed that regular people can manifest their artistic ideas without needing to go through an artist or spend years studying the craft. The artists are calling the AI companies oppressors because they are breaking the artist's stranglehold on the market.
It's incredibly ironic how socializing what was a privatized ability has otherwise "socialist" people completely losing their shit. Just the mask of pure virtue slipping...
Software development was already one of the most democratised professions on earth. With any old dirt cheap used computer, an internet connection, and enough drive and curiosity you could self-train yourself into a role that could quickly become a high paying job. While they certainly helped, you never needed any formal education or expensive qualifications to excel in this field. How is this better?
Maybe not as good, but they can certainly do far far more than what was available a few years ago.
That's primarily why I finally had to suck it up and sign up for Claude. Claude clearly can cough up proprietary codebase examples that I otherwise have no access to.
As far as I can tell the reason OpenAI and Anthropic are ahead in code is that they've invested extremely heavily in figuring out the right reinforcement learning training mix needed to get great coding results.
Some of the Chinese open models are already showing signs of catching up.
> simonw: It's better because now you can automate something tedious in your life with a computer without having to first climb a six month learning curve.
Completely ignores, or enthusiastically accepts and endorses, the consolidation of production, power, and wealth into a stark few (friends), and claims superiority and increased productivity without evidence?
This may be the most simonw comment I have ever seen.
I'm not worried about that at all any more. There are dozens of organizations who have achieved that milestone now, and OpenAI aren't even definitively in the lead.
A lot of those top-class models are open weight (mainly thanks to the Chinese labs) and available for people to run on their own hardware.
I wrote a bunch more about this in my 2024 wrap-up: https://simonwillison.net/2024/Dec/31/llms-in-2024/#the-gpt-...
She used to log into 3 different websites. Now she just opens localhost:3000 and has all of them on the same page. No emails shared with anyone. All data stored locally.
I could have done this earlier but the time commitment with Claude Code now was writing a spec in 5-minutes and pressing approve a few times vs half a day.
I count this as an absolute win. No privacy breaches, no data sharing.
Ascribing the goals of "democratize code/art" onto these companies and their products is called delusion.
I am sure the 3 letter agency directors on these company boards are thrilled you think they left their lifelong careers solely to finally realize their dream to allow you to code and "manifest your artistic ideas".
So maybe celebrate open/private/local models for empowering people rather than selfishly complain about it?
It's more economical for the average person to spend $20/month on a subscription than it is for them to drop multiple thousands $ and untold hours of time experimenting. Local AI is a fun hobby though.
It's incredibly ironic how blatant theft has left otherwise capitalistic people so enthusiastic.
Tt's because these companies profit from all the existing art without compensating the artists. Even worse, they are now putting the very people out of a job who (unwittingly) helped to create these tools in the first place. Not to mention how hurtful it must be for artists seeing their personal style imitated by a machine without their consent.
I totally see how it can empower regular people, but it also empowers the megacorps and bad actors. The jury is still out on whether AI is providing a net positive to society. Until then, let's not ignore the injustice and harm that went into creating these tools and the potential and real dangers that come with it.
The vast majority of coders in employment barely write anything more complex than basic CRUD apps. These jobs were always going to be automated or abstracted away sooner or later.
Every profession changes. Saying that these new tools are useless or won't impact you/xyz devs is just ignoring a repeated historical pattern
Similarly, Lisp's homoiconicity makes sense to me as a wonderfully aesthetic idea. I remember generating strings-of-text that were code, but still just text, and wishing that I could trivially step into the structure there like it was a map/dict... without realizing that that's what an AST is and what the language compiler / runtime is already always doing.
[1] Like the recent "Gas Town" and "Beads" that people keep mentioning in the comments that require extensive scripts/human intervention to purge from the system: https://news.ycombinator.com/item?id=46510121
My experience has been negative progress in this field. On iOS, UIKit in Interface Builder is an order of magnitude faster to write and to debug, with less weird edge cases, than SwiftUI was last summer. I say last summer because I've been less and less interested in iOS the more I learn about liquid glass, even ignoring the whole "aaaaaaa" factor of "has AI made front end irrelevant anyway?" and "can someone please suggest something the AI really can't do so I can get a job in that?"
Too bad they were mostly proprietary and won't even run in modern hardware.
And it's actually kind of amazing, because a lot of people who earn six figures programming Salesforce came to it from a non-traditional software engineering background.
I even gave -True Vibe Coding- a whirl. Yesterday, from a blank directory and text file list of requirements, I had Opus 4.5 build an Android TV video player that could read a directory over NFS, show a grid view of movie poster thumbnails, and play the selected video file on the TV. The result wasn't exactly full-featured Kodi, but it works in the emulator and actual device, it has no memory leaks, crashes, ANRs, no performance problems, no network latency bugs or anything. It was pretty astounding.
Oh, and I did this all without ever opening a single source file or even looking at the proposed code changes while Opus was doing its thing. I don't even know Kotlin and still don't know it.
At work I only have access to calude using the GitHub copilot integration so this could be the cause of my problems. Claude was able to get slthe first iteration up pretty quick. At that stage the app could create a plot and you could interact with it and ask follow up questions.
Then I asked it to extend the app so that it could generate multiple plots and the user could interact with all of them one at a time. It made a bunch of changes but the feature was never implemented. I asked it to do again but got the same outcome. I completely accept the fact that it could just be all because I am using vscode copilot or my promoting skills are not good but the LLM got 70% of the way there and then completely failed
You really need to at least try Claude Code directly instead of using CoPilot. My work gives us access to CoPilot, Claude Code, and Codex. CoPilot isn’t close to the other more agentic products.
For instance, I always respected types, but I'm too lazy to go spend hours working on types when I can just do ruby-style duck typing and get a long ways before the inevitable problems rear their head. Now, I can use a strongly typed language and get the advantages for "free".
I find I'm doing more Typescript projects than Python because of the superior typing, despite the fact I prefer Python.
Oh absolutely. I've been using Python for past 15 or so years for everything.
I've never written a single line of Rust in my life, and all my new projects are Rust now, even the quick-script-throwaway things, because it's so much better at instantly screaming at claude when it goes off track. It may take it longer to finish what I asked it to do, but requires so much less involvement from me.
I will likely never start another new project in python ever.
EDIT: Forgot to add that paired with a good linter, this is even more impressive. I told Claude to come up with the most masochistic clippy configuration possible, where even a tiny mistake is instantly punished and exceptions have to be truly exceptional (I have another agent that verifies this each run).
I just wish there was cargo-clippy for enforcing architectural patterns.
Going to one-up you though -- here's a literal one-liner that gets me a polished media center with beautiful interface and powerful skinning engine. It supports Android, BSD, Linux, macOS, iOS, tvOS and Windows.
`git clone https://github.com/xbmc/xbmc.git`
A big part of the problem with existing software is that humans seem to be pretty much incapable of deciding a project is done and stop adding to it. We treat creating code like a job or hobby instead of a tool. Nothing wrong with that, unless you're advertising it as a tool.
The old adage about how "users use 10% of your software's features, but they each use a different 10%" can now be solved by each user just building that 10% for themselves.
If you were someone saying at the start of 2025 "this is a flash in the pan and a bunch of hype, it's not going to fundamentally change how we write code", that was still a reasonable belief to hold back then. At the start of 2026 that position is basically untenable: it's just burying your head in the sand and wishing for AI to go away. If you're someone who still holds it you really really need to download Claude Code and set it to Opus and start trying it with an open mind: I don't know what else to tell you. So now the question has shifted from whether this is going to transform our profession (it is), to how exactly it's going to play out. I personally don't think we will be replacing human engineers anytime soon ("coders", maybe!), but I'm prepared to change my mind on that too if the facts change. We'll see.
I was a fellow mind-changer, although it was back around the first half of last year when Claude Code was good enough to do things for me in a mature codebase under supervision. It clearly still had a long way to go but it was at that tipping point from "not really useful" to "useful". But Opus 4.5 is something different - I don't feel I have to keep pulling it back on track in quite the way I used to with Sonnet 3.7, 4, even Sonnet 4.5.
For the record, I still think we're in a bubble. AI companies are overvalued. But that's a separate question from whether this is going to change the software development profession.
I'm hoping after AI comes back down to earth there will be a new glut of cheap second hand GPUs and RAM to get snapped up.
I think my comparison is apt; being a bubble and a truly society-altering technology are not mutually exclusive, and by virtue of it being a bubble, it is overhyped.
but I do want a better way to glance and keep up with what its doing in longer conversations, for my own mental context window
I'm finding that branding and graphic design is the most arduous part, that I'm hoping to accelerate soon. I'm heavily AI assisted there too and I'm evaluating MCP servers to help, but so far I do actually have to focus on just that part as opposed to babysit
... says it all.
Literally tried it yesterday. I didn't see a single difference with whatever model Claude Code was using two months ago. Same crippled context window. Same "I'll read 10 irrelevant lines from a file", same random changes etc.
Create a markdown document of your task (or use CLAUDE.md), put it in "plan mode" which allows Claude to use tool calls to ask questions before it generates the plan.
When it finishes one part of the plan, have it create a another markdown document - "progress.md" or whatever with the whole plan and what is completed at that point.
Type /clear (no more context window), tell Claude to read the two documents.
Repeat until even a massive project is complete - with those 2 markdown documents and no context window issues.
... Proceeds to explain how it's crippled and all the workarounds you have to do to make it less crippled.
No - that's not what I did.
You don't need an extra-long context full of irrelevant tokens. Claude doesn't need to see the code it implemented 40 steps ago in a working method from Phase 1 if it is on Phase 3 and not using that method. It doesn't need reasoning traces for things it already "thought" through.
This other information is cluttering, not helpful. It is making signal to noise ratio worse.
If Claude needs to know something it did in Phase 1 for Phase 4 it will put a note on it in the living markdown document to simply find it again when it needs it.
What I very succinctly called "crippled context" despite claims that Opus 4.5 is somehow "next tier". It's all the same techniques we've been using for over a year now.
And even my short-term memory is significantly larger than the at most 50% of the 200k-token context window that Claude has. It runs out of context before my short-term memory is probably not even 1% full, for the same task (and I'm capable of more context-switching in the meantime).
And so even the "Opus 4.5 really is at a new tier" runs into the very same limitations all models have been running into since the beginning.
For LLMs long term memory is achieved by tooling. Which you discounted in your previous comments.
You also overstimate capacity of your short-term memory by few orders of magnitude:
https://my.clevelandclinic.org/health/articles/short-term-me...
My specific complaint, which is an observable fact about "Opus 4.5 is next tier": it has the same crippled context that degrades the quality of the model as soon as it fills 50%.
EMM_386: no-no-no, it's not crippled. All you have to do is keep track across multiple files, clear out context often, feed very specific information not to overflow context.
Me: so... it's crippled, and you need multiple workarounds
scotty79: After all it's the same as your own short-term memory, and <some unspecified tooling (I guess those same files)> provide long-term memory for LLMs.
Me: Your comparison is invalid because I can go have lunch, and come back to the problem at hand and continue where I left off. "Next tier Opus 4.5" will have to be fed the entire world from scratch after a context clear/compact/in a new session.
Unless, of course, you meant to say that "next tier Opus model" only has 15-30 second short term memory, and needs to keep multiple notes around like the guy from Memento. Which... makes it crippled.
Who said I refuse them?
I evaluated the claim that Opus is somehow next tier/something different/amazeballs future at its face value. It still has all the same issues and needs all the same workarounds as whatever I was using two months ago (I had a bit of a coding hiatus between beginning of December and now).
> then you end up with a guy from Memento and regardless of how smart the model is
Those models are, and keep being the guy from memento. Your "long memory" is nothing but notes scribbled everywhere that you have to re-assemble every time.
> And that's why you can't tell the difference between smarter and dumber one while others can.
If it was "next tier smarter" it wouldn't need the exact same workarounds as the "dumber" models. You wouldn't compare the context to the 15-30 second short-term memory and need unspecified tools [1] to have "long-term memory". You wouldn't have the model behave in an indistinguishable way from a "dumber" model after half of its context windows has been filled. You wouldn't even think about context windows. And yet here we are
[1] For each person these tools will be a different collection of magic incantations. From scattered .md files to slop like Beads to MCP servers providing access to various external storage solutions to custom shell scripts to ...
BTW, I still find "superpowers" from https://github.com/obra/superpowers to be the single best improvement to Claude (and other providers) even if it's just another in a long serious of magic chants I've evaluated.
That's exactly how the long term memory works in humans as well. The fact that some of these scribbles are done chemically in the same organ that does the processing doesn't make it much better. Human memories are reassembled at recall (often inaccurately). And humans also scribble when they try to solve a problem that exceeds their short term memory.
> If it was "next tier smarter" it wouldn't need the exact same workarounds as the "dumber" models.
This is akin to opposing calling processor next tier because it still needs RAM and bus to communicate with it and SSD as well. You think it should have everything in cache to be worthy of calling it next tier.
It's fine to have your own standards for applying words. But expect further confusion and miscommunication with other people if don't intend to realign.
Where this is applicable when you go away from a problem for a while. And yet I don't lose the entire context and have to rebuild it from scratch when I go for lunch, for example.
Models have to rebuild the entire world from scratch for every small task.
> This is akin to opposing calling processor next tier because it still needs RAM and bus to communicate with it and SSD as well.
You're so lost in your own metaphor that it makes no sense.
> You think it should have everything in cache to be worthy of calling it next tier.
No. "Next tier" implies something significantly and observably better. I don't. And here you are trying to tell me "if you use all the exact same tools that you have already used before with 'previous tier models' you will see it is somehow next tier".
If your "next tier" needs an equator-length list of caveats and all the same tools, it's not next tier is it?
BTW. I'm literally coding with this "next tier" tool with "long memory just like people". After just doing the "plan/execute/write notes" bullshit incantations I had to correct it:
You're right, I fucked up on all three counts:
1. FileDetails - I should have WIRED IT UP, not deleted it.
It's a useful feature to preview file details before playing.
I treated "unused" as "unwanted" instead of "not yet connected".
2. Worktree not merged - Complete oversight. Did all the work but
didn't finish the job.
3. _spacing - Lazy fix. Should have analyzed why it exists and either
used it or removed the layout constraint entirely.
Oh. Within about 10 seconds after that it started compacting the "non-crippled" context window and immediately forgot most of what it had just been doing. So I had to clear out the context and teach it the world from the start again.
> It's fine to have your own standards for applying words. But expect further confusion and miscommunication with other people if don't intend to realign.
I mean, just like crypto bros before them, AI bros do sure love to invent their own terminology and their own realities that have nothing to do with anything real and observable.
It very well might be that AI tools are not for you, if you are getting such poor results with your methods of approaching them.
If you would like to improve your outcomes at some point, ask people who achieve better results for pointers and try them out. Here's a freebie, never tell AI it fucked up.
Yup. There's some magical "right context" that will fix all the problems. What is that right context? No idea, I guess I need to read a yet-another 20 000-word post describing magical incantations that you should or shouldn't do in the context.
The "Opus 4.5 is something else/nex tier/just works" claims in my mind means that I wouldn't need to babysit its every decision, or that it would actually read relevant lines from relevant files etc. Nope. Exact same behaviors as whatever the previous model was.
Oh, and that "200k tokens context window"? It's a lie. The quality quickly degrades as soon as Claude reaches somewhere around 50% of the context window. At 80+% it's nearly indistinguishable from a model from two years ago. (BTW, same for Codex/GPT with it's "1 million token window")
1) define problem
2) split problem into small independently verifiable tasks
3) implement tasks one by one, verify with tools
With an LLM usually 1) is just a markdown file, 2) is a markdown checklist, Github issues (which Claude can use with the `gh` cli) and every loop of 3 gets a fresh context, maybe the spec from step 1 and the relevant task information from 2
I haven't ran into context issues in a LONG time, and if I have it's usually been either intentional (it's a problem where compacting wont' hurt) or an error on my part.
> I haven't ran into context issues in a LONG time
Because you've become the reverse centaur :) "a person who is serving as a squishy meat appendage for an uncaring machine." [1]
You are very aware of the exact issues I'm talking about, and have trained yourself to do all the mechanical dance moves to avoid them.
I do the same dances, that's why I'm pointing out that they are still necessary despite the claims of how model X/Y/Z are "next tier".
[1] https://doctorow.medium.com/https-pluralistic-net-2025-12-05...
The exact same method that worked for those happened to also work for LLMs, I didn't have to learn anything new or change much in my workflow.
"Fix bug in FoobarComponent" is enough of a bug ticket for the 100x developer in your team with experience with that specific product, but bad for AI, juniors and offshored teams.
Thus, giving enough context in each ticket to tell whoever is working on it where to look and a few ideas what might be the root cause and how to fix it is kinda second nature to me.
Also my own brain is mostly neurospicy mush, so _I_ need to write the context to the tickets even if I'm the one on it a few weeks from now. Because now-me remembers things, two-weeks-from-now me most likely doesn't.
It's in your interest to deal with your frustration and figure out how you can leverage the new tools to stay relevant (to the degree that you want to).
Regarding the context window, Claude needs thinking turned up for long context accuracy, it's quite forgetful without thinking.
No one attacked your use of AI. I explained my own experience with the "Claude Opus 4.5 is next tier". You barged in, ignored anything I said, and attacked my skills.
> the workplace is going to punish people who don't leverage AI though, and I'm trying to be helpful.
So what exactly is helpful in your comments?
(And it's fine to do so, just don't mail bombs to us, ok?)
As for the snide and patronizing "it's in your interest to stay relevant":
1. I use these tools daily. That's why I don't subscribe to willful wide-eyed gullibility. I know exactly what these tools can and cannot do.
The vast majority of "AI skeptics" are the same.
2. In a few years when the world is awash in barely working incomprehensible AI slop my skills will be in great demand. Not because I'm an amazing developer (I'm not), but because I have experience separating wheat from the chaff
It seems the subject of AI is emotionally charged for you, so I expect friendly/rational discourse is going to be a challenge. I'd say something nice but since you're primed to see me being patronizing... Fuck you? That what you were expecting?
It's not me who decided to barge in, assume their opponent doesn't use something or doesn't want to use something, and offer unsolicited advice.
> It kinda makes me sad when the discourse is so poisoned that I can't even encourage someone to protect their own future from something that's obviously coming
See. Again. You're so in love with your "wisdom" that you can't even see what you sound like: snide, patronising, condenscending. And completely missing the whole point of what was written. You are literally the person who poisons the discourse.
Me: "here are the issues I still experience with what people claim are 'next tier frontier model'"
You: "it's in your interests to figure out how to leverage new tools to stay relevant in the future"
Me: ... what the hell are you talking about? I'm using these tools daily. Do you have anything constructive to add to the discourse?
> so I expect friendly/rational discourse is going to be a challenge.
It's only challenge to you because you keep being in love with your voice and your voice only. Do you have anything to contribute to the actual rational discourse, are you going to attack my character?
> 'd say something nice but since you're primed to see me being patronizing... Fuck you? T
Ah. The famous friendly/rational discourse of "they attack my use of AI" (no one attacked you), "why don't you invest in learning tools to stay relevant in the future" (I literally use these tools daily, do you have anything useful to say?) and "fuck you" (well, same to you).
> That what you were expecting?
What I was expecting is responses to what I wrote, not you riding in on a high horse.
If you want to take politeness as being patronizing, I'm happy to stop bothering. My guess is you're not a special snowflake, and you need to "get good" or you're going to end up on unemployment complaining about how unfair life is. I'd have sympathy but you don't seem like a pleasant human being to interact with, so have fun!
They are not giving me the results people claim they give. It is distinctly different from not giving the results I want.
> If you're using these tools daily and having a hard time, either you're working on something very different from the bulk of people using the tools and your problems or legitimate, or you aren't and it's a skill issue.
Indeed. And your rational/friendly discourse that you claim you're having would start with trying to figure that out. Did you? No, you didn't. You immediately assumed your opponent is a clueless idiot who is somehow against AI and is incapable or learning or something.
> If you want to take politeness as being patronizing, I'm happy to stop bothering.
No. It's not politeness. It's smugness. You literally started your interaction in this thread with a "git gud or else" and even managed to complain later that "you dislike it when they attack your use of AI as a skill issue". While continuously attacking others.
> you don't seem like a pleasant human being to interact with
Says the person who has contributed nothing to the conversation except his arrogance, smugness, holier-than-thou attitude, engaged in nothing but personal attacks, complained about non-existent grievances and when called out on this behavior completed his "friendly and rational discourse" with a "fuck you".
Well, fuck you, too.
Adieu.
All I can tell you is that in my own lived experience, I've had some fantastic results from AI, and it comes from telling it "look at this thing here, ok, i want you to chain it to that, please consider this factor, don't forget that... blah blah blah" like how I would have spelled things out to a junior developer, and then it really does stand a really solid chance of turning out what I've asked for. It helps a lot that I know what to ask for; there's no replacing that with AI yet.
So, your own situation must fall into one of these coarse buckets:
- You're doing something way too hard for AI to have a chance at yet, like real science / engineering at the frontier, not just boring software or infra development
- Your prompts aren't specific enough, you're not feeding it context, and you're expecting it to one-shot things perfectly instead of having to spend an afternoon prompting and correcting stuff
- You're not actually using and getting better at the tools, so you're just shouting criticisms from the sidelines, perhaps as sour grape because you're not allowed by policy / company can't afford to have you get into it.
IDK. I hope it's the first one and you're just doing Really Hard Things, but if you're doing normal software developer stuff and not seeing a productivity advantage, it's a fucking skill issue.
Are you sure you're not an LLM?
2. It's the same workarounds we've been doing forever
3. It's indistinguishable from "clear context and re-feed the entire world of relevant info from scratch" we've had forever, just slightly more automated
That's why I don't understand all the "it's new tier" etc. It's all the same issues with all the same workarounds.
It's not just the deficiencies of earlier versions, but the mismatch between the praise from AI enthusiasts and the reality.
I mean maybe it is really different now and I should definitely try uploading all of my employer's IP on Claude's cloud and see how well it works. But so many people were as hyped by GPT-4 as they are now, despite GPT-4 actually being underwhelming.
Too much hype for disappointing results leads to skepticism later on, even when the product has improved.
The funny part about rapidly changing industries is that, despite the fomo, there's honestly not any reward to keeping up unless you want to be a consultant. Otherwise, wait and see what sticks. If this summer people are still citing the Opus 4.5 was a game changing moment and have solid, repeatable workflows, then I'll happily change up my workflow.
Someone could walk into the LLM space today and wouldn't be significantly at a loss for not having paid attention to anything that had happened in the last 4 years other than learning what has stuck since then.
LMAO what???
Two big examples:
- Period from early mvc JavaScript frontends (backbone.js etc) and the time of the great React/Angular wars. I completely stepped out of the webdev space during that time period.
- The rapid expansion of Deep Learning frameworks where I did try to keep up (shipped some Lua torch packages and made minor contributions to Pylearn2).
In the first case, missing 5 years of front-end wars had zero impact. After not doing webdev work at all for 5-years I was tasked with shipping a React app. It took me a week to catch up, and everything was deployed in roughly the same time as someone would have had they spent years keeping up with changes.
In the second case, where I did keep up with many of the developing deep learning frameworks, it didn't really confer any advantage. Coworkers who I worked with who started with Pytorch fresh out of school were just as proficient, if not more so, with building models. Spending energy keeping up offered no value other than feeling "current" at the time.
Can you give me a counter example of where keeping up with a rapidly changing area that's unstable has conferred a benefit to you? Most of FOMO is really just fear. Again, unless you're trying to sell your self specifically as a consultant on the bleeding edge, there's no reason to keep up with all these changes (other than finding it fun).
Right or wrong - doesn’t matter. You typed in a line of text and now your computer is making 3000 word stories, images, even videos based on it
How are you NOT astounded by that? We used to have NONE of this even 4 years ago!
-- Charles Babbage
My team is using it with Claude Code and say it works brilliantly, so I'll be giving it another go.
How much of the value comes from Opus 4.5, how much comes from Claude Code, and how much comes from the combination?
Copilot has by far the best and most intuitive agent UI. Just make sure you're in agent mode and choose Sonnet or Opus models.
I've just cancelled my Claude sub and gone back and will upgrade to the GH Pro+ to get more sonnet/opus.
The gap didn't seem big, but in November (which admittedly was when Opus 4.5 was in preview on Copilot) Opus 4.5 with Copilot was awful.
This is what people are still doing wrong. Tools in a loop people, tools in a loop.
The agent has to have the tools to detect whatever it just created is producing errors during linting/testing/running. When it can do that, I can loop again, fix the error and again - use the tools to see whether it worked.
I _still_ encounter people who think "AI programming" is pasting stuff into ChatGPT on the browser and they complain it hallucinates functions and produces invalid code.
Well, d'oh.
I was having Opus investigate it and I kept building and deploying the firmware for testing.. then I just figured I'd explain how it could do the same and pull the logs.
Off it went, for the next ~15 minutes it would flash the firmware multiple times until it figured out the issue and fixed it.
There was something so interesting about seeing a microcontroller on the desk being flashed by Claude Code, with LEDs blinking indicating failure states. There's something about it not being just code on your laptop that felt so interesting to me.
But I agree, absolutely, red/green test or have a way of validating (linting, testing, whatever it is) and explain the end-to-end loop, then the agent is able to work much faster without being blocked by you multiple times along the way.
While Claude is amazing at writing code, it still requires human operators. And even experienced human operators are bad at operating this machinery.
Tell your average joe - the one who thinks they can create software without engineers - what "tools-in-a-loop" means, and they'll make the same face they made when you tried explaining iterators to them, before LLMs.
Explain to them how typing system, E2E or integration test helps the agent, and suddenly, they now have to learn all the things they would be required to learn to be able to write on their own.
I had an idea for something that I wanted, and in five scattered hours, I got it good enough to use. I'm thinking about it in a few different ways:
1. I estimate I could have done it without AI with 2 weeks full-time effort. (Full-time defined as >> 40 hours / week.)
2. I have too many other things to do that are purportedly more important that programming. I really can't dedicate to two weeks full-time to a "nice to have" project. So, without AI, I wouldn't have done it at all.
3. I could hire someone to do it for me. At the university, those are students. From experience with lots of advising, a top-tier undergraduate student could have achieved the same thing, had they worked full tilt for a semester (before LLMs). This of course assumes that I'm meeting them every week.
FYI, I still use cursor for small edits and reviews.
Seriously, I have 3+ claude code windows open at a time. Most days I don't even look at the IDE. It's still there running in the background, but I don't need to touch it.
- single scripts. Anything which can be reduced to a single script.
- starting greenfield projects from scratch
- code maintenance (package upgrades, old code...)
- tasks which have a very clear and single definition. This isn't linked to complexity, some tasks can be both very complex but with a single definition.
If your work falls into this list they will do some amazing work (and yours clearly fits that), if it doesn't though, prepare yourself because it will be painful.
I should say I'm hardly ever vibe-coding, unlike the original article. If I think I want code that will last, I'll steer the models in ways that lean on years of non-LLM experience. E.g., I'll reject results that might work if they violate my taste in code.
It also helps that I can read code very fast. I estimate I can read code 100x faster than most students. I'm not sure there is any way to teach that other than the old-fashioned way, which involves reading (and writing) a lot of code.
Yes indeed, these are the things on the other hand which aren't working well in my opinion:
- large codebase
- complex domain knowledge
- creating any feature where you need product insights
- tasks requiring choices (again, complexity doesn't matter here, the task may be simple but require some choices)
- anything unclear where you don't know where you are going first
While you don't experience any of these when teaching or side projects, these are very common in any enterprise context.
I then proceeded to use it to hack on its own codebase, and close a bunch of issues in a repository that I maintain (https://github.com/nuprl/MultiPL-E/commits/main/).
> have it learn your conventions, pull in best practices
What do you mean by "have it learn your conventions"? Is there a way to somehow automatically extract your conventions and store it within CLAUDE.md?
> For example, we have a custom UI library, and Claude Code has a skill that explains exactly how to use it. Same for how we write Storybooks, how we structure APIs, and basically how we want everything done in our repo. So when it generates code, it already matches our patterns and standards out of the box.
Did you have to develop these skills yourself? How much work was that? Do you have public examples somewhere?
From the docs (https://hexdocs.pm/ash/what-is-ash.html):
"Model your application's behavior first, as data, and derive everything else automatically. Ash resources center around actions that represent domain logic."
I'll give you an example: I use ruff to format my python code, which has an opinionated way of formatting certain things. After an initial formatting, Opus 4.5, without prompting, will write code in this same style so that the ruff formatter almost never has anything to do on new commits. Sonnet 4.5 is actually pretty good at this too.
I care about the norms in my codebase that can't be automatically enforced by machine. How is state managed? How are end-to-end tests written to minimize change detectors? When is it appropriate to log something?
We have some tests in "GIVEN WHEN THEN" style, and others in other styles. Opus will try to match each style of testing by the project it is in by reading adjacent tests.
"AI has X"
"We have X at home"
"X at home: x"
But I think it should be doable. You can tell it how YOU want the state to be managed and then have it write a custom "linter" that makes the check deterministic. I haven't tried this myself, but claude did create some custom clippy scripts in rust when I wanted to enforce something that isn't automatically enforced by anything out there.
I’ve even found it searching node_modules to find the API of non-public libraries.
Claude Code will spawn sub-agents (that often use their cheap Haiki model) for exploration and planning tasks, with only the results imported into the main context.
I've found the best results from a more interactive collaboration with Claude Code. As long as you describe the problem clearly, it does a good job on small/moderate tasks. I generally set two instances of Claude Code separate tasks and run them concurrently (the interaction with Claude Code distracts me too much to do my own independent coding simultaneously like with setting a task for a colleague, but I do work on architecture / planning tasks)
The one manner of taste that I have had to compromise on is the sheer amount of code - it likes to write a lot of code. I have a better experience if I sweat the low-level code less, and just periodically have it clean up areas where I think it's written too much / too repetitive code.
As you give it more freedom it's more prone to failure (and can often get itself stuck in a fruitless spiral) - however as you use it more you get a sense of what it can do independently and what's likely to choke on. A codebase with good human-designed unit & playwright tests is very good.
Crucially, you get the best results where your tasks are complex but on the menial side of the spectrum - it can pay attention to a lot of details, but on the whole don't expect it to do great on senior-level tasks.
To give you an idea, in a little over a month "npx ccusage" shows that via my Claude Code 5x sub I've used 5M input tokens, 1.5M output, 121M Cache Create, 1.7B Cache Read. Estimated pay-as-you-go API cost equivalent is $1500 (N.B. for the tail end of December they doubled everybody's API limits, so I was using a lot more tokens on more experimental on-the-fly tool construction work)
I exclusively use opus for architecture / speccing, and then mostly Sonnet and occasionally Haiku to write the code. If my usage has been light and the code isn't too straightforward, I'll have Opus write code as well.
I have hit the weekly limit before, briefly, but that took running multiple sessions in parallel continuously for many days.
/init in Claude Code already automatically extracts a bunch, but for something more comprehensive, just tell it which additional types of things you want it to look for and document.
> Did you have to develop these skills yourself? How much work was that? Do you have public examples somewhere?
I don't know about the person above, but I tell Claude to write all my skills and agents for me. With some caveats, you can do this iteratively in a single session ("update the X agent, then re-run it. Repeat until it reliably does Y")
Until such time, of course, when LLMs are eating their own dogfood, in which case they - as has already happened - create their own language, evolve dramatically, and cue skynet.
Likewise, many optimization techniques involve some randomness, whether it's approximating an NP-thorny subproblem, or using PGO guided by statistical sampling. People might disable those in pursuit of reproducible builds, but no one would claim that enabling those features makes GCC or LLVM no longer a compiler. So nondeterminism isn't really the distinguishing factor either.
codex cli even has a skill to create skills; it's super easy to get up to speed with them
https://github.com/openai/skills/blob/main/skills/.system/sk...
Why on earth would you be hunting $20 a month subscriptions from random assed people? Peanuts.
Lockheed-Martin could be, but isn’t, opening lemonade stands outside their offices… they don’t because of how buying a Ferrari works.
For the same reason Microsoft never has and never will chase people for pirating home Windows or Office licenses
When they hit the workforce, or even better, start a company guess which OS and office suite they'll use? Hint: It's not Linux and Openoffice.
Same with Claude's $20 package. It lets devs use it at home and then compare it to the Copilot shit their company is pushing on them. Maybe they either grumble enough to get a Claude license or they're in a position to make the call.
Cheap advertising pretty much.
Worked for me too :) I've paid my own Claude license for over a year at home, grumbled at work and we got a Claude pilot going now - and everyone who's tried it so far isn't going back to Copilot + Sonnet 4.5/GPT5.
LLMs do not result in bosses firing people, it results in more projects / faster completed projects, what in turn means more $$$ for a company.
Well excuse the shit out of my goddamn French, but being comfy for years and suddenly facing literal doom of my profession in a year wasn't on my bingo card.
And what do you even mean by "prepare"? Shit out a couple of mil out of my ass and invest asap?
Not the person you're responding to but... if you think it's a horse -> car change (and, to stretch the metaphor, if you think you're in the business of building stables) then preparation means train in another profession.
If you think it's a hand tools -> power tools change, learn how to use the new tools so you don't get left behind.
My opinion is it's a hand -> power tools change, and that LLMs give me the power to solve more problems for clients, and do it faster and more predictably than a client trying to achieve the same with an LLM. I hope I'm right :-)
You think any manager in their right mind would take risks like that?
I think the real consequences are that they probably are so pleased with how productive the team is becoming that they will not hire new people or fire the ones who aren't keeping up with the times.
It's like saying "wow, our factory just produced 50% more cars this year, time to shut down half the factory to reduce costs!"
You really underestimate stupidity of your average manager. Two of our top performers left because they were underpaid and the manager (in charge of the comp) never even tried to retain them.
Ignoring that this same developer, now has access to a tool, that makes himself a team.
Going independent was always a issue because being a full stack dev, is hard. With LLMs, you have a entire team behind you for making graphics, code, documents, etc... YOU becomes the manager.
We will see probably a lot more smaller teams/single devs making bigger projects, until they grow.
The companies that think they can fire devs, are the same companies that are going to go too far, and burn bridges. Do not forget that a lot of companies are founded on devs leaving a company, and starting out on their own, taking clients with them!
I did that years ago, and it worked for a while but eventually the math does not work out because one guy can only do so much. And when you start hiring, your costs balloon. But with LLMs ... Now your a one man team, ... hiring a second person is not hiring a person to make some graphics or doing more coding. Your hiring another team.
This is what people do not realize... they look too much upon this as the established order, ignoring what those fired devs now can do!
Do you really want to be a middle manager to a bunch of text boxes, churning out slop, while they drive up our power bills and slowly terraform the planet?
Ironically, that race to the bottom is no different then we already have. Have you already worked for a company before? A lot of software is developed, BADLY. I dare to say that a lot of software that Opus 4.5 generates, is often a higher quality then what i have seen in my 25 year carrier.
The amount of companies that cheapen out, hiring juniors fresh from school, to work as coding monkies is insane. Then projects have bugs / security issues, with tons of copy/pasted code, or people not knowing a darn thing.
Is that any different then your feared future? I dare to say, that LLms like Opus are frankly better then most juniors. As a junior to do a code review for security issues. Opus literally creates extensive tests, points out issues that you expect from a mid or higher level dev. Of course, you need to know to ask! You are the manager.
> Do you really want to be a middle manager to a bunch of text boxes, churning out slop, while they drive up our power bills and slowly terraform the planet?
Frankly, yes ... If you are a real developer, do you still think development is fun after 10 years, 20 years? Doing the exact same boring work. Reimplementing the 1001 login page, the 101 contact form ... A ton of our work is in reality repeating the same crap over and over again. And if we try to bypass it, we end up tied to tied to those systems / frameworks that often become a block around our necks.
Our industry has a lot of burnout because most tasks may start small but then grow beyond our scope. Todays its ruby on rails programming, then its angular, no wait, react, no wait, Vue, no wait, the new hotness is whatever again.
> slowly terraform the planet?
Well, i am actually making something.
Can you say the same for all the power / gpu draw with bitcoin, Ethereum whatever crap mining. One is productive, a tool with insane potential and usage, the other is a virtual currency where only one is ever popular with limited usage. Yet, it burns just as much for a way more limited return of usability.
Those LLMs that you are so against, make me a ton more productive. You wan to to try out something, but never really wanted to get committed because it was weeks of programming. Well, now you as manager, can get projects done fast. Learn from them way faster then your little fingers ever did.
Just like farmers who couldn't cope with the additional leverage their equipment provided them, devs who can't leverage this technology will have to "go to the cities".
If LLM's turn out the way C-Suite hopes. Let me tell you, you will be in a world of pain. Most of you won't be using LLM's to create your own businesses.
Seeing the replies here it actually doesn't seem like everyone else can do this. Looks like a lot of people really suck at using LLMs to me.
Disclaimer: I don't have access to Claude Code. My employer has only granted me Claude Teams. Supposedly, they don't use my poopy code to train their models if I use my work email Claude so I am supposed to use that. If I'm not pasting code (asking general questions) into Claude, I believe I'm allowed to use whatever.
pardon my ignorance.
I use GitHub Copilot which has access to llms like Gemini 3, Sonnet/Opus 4.5 ang GPT 5.2
Key word: seems. It's impossible to do a proper qualitative analysis.
When needed it can be picked up in a day. Otherwise they are not paid based in tickets solved etc. If the incentives were properly aligned everyone would already use it
Reword? But why not just voice to text alone...
Oh but we all read the partially synthetic ad by this point. Psyche.
Nobody is sleeping. I'm using LLMs daily to help me in simple coding tasks.
But really where is the hurry? At this point not a few weeks go by without the next best thing since sliced bread to come out. Why would I bother "learning" (and there's really nothing to learn here) some tool/workflow that is already outdated by the time it comes out?
> 2026 is going to be a wake-up call
Do you honestly think a developer not using AI won't be able to adapt to a LLM workflow in, say, 2028 or 2029? It has to be 2026 or... What exactly?
There is literally no hurry.
You're using the equivalent of the first portable CD-player in the 80s: it was huge, clunky, had hiccups, had a huge battery attached to it. It was shiny though, for those who find new things shiny. Others are waiting for a portable CD player that is slim, that buffers, that works fine. And you're saying that people won't be able to learn how to put a CD in a slim CD player because they didn't use a clunky one first.
It's also a skill that compounds over time, so if you have two years of experience with them you'll be able to use them more effectively than someone with two months of experience.
In that respect, they're just normal technology. A Python programmer with two years of Python experience will be more effective than a programmer with two months of Python.
That is sleeping.
> But really where is the hurry? At this point not a few weeks go by without the next best thing since sliced bread to come out. Why would I bother "learning" (and there's really nothing to learn here) some tool/workflow that is already outdated by the time it comes out?
You're jumping to conclusions that haven't been justified by any of the development in this space. The learning compounds.
> Do you honestly think a developer not using AI won't be able to adapt to a LLM workflow in, say, 2028 or 2029? It has to be 2026 or... What exactly?
They will, but they'll be competing against people with 2-3 more years of experience in understanding how to leverage these tools.
Codex/Claude Code are terrible with C++. It also can't do Rust really well, once you get to the meat of it. Not sure why that is, but they just spit out nonsense that creates more work than it helps me. It also can't one shot anything complete, even though I might feed him the entire paper that explains what the algorithm is supposed to do.
Try to do some OpenGL or Vulkan with it, without using WebGPU or three.js. Try it with real code, that all of us have to deal with every day. SDL, Vulkan RHI, NVRHI. Very frustrating.
Try it with boost, or cmake, or taskflow. It loses itself constantly, hallucinates which version it is working on and ignores you when you provide actual pointers to documentation on the repo.
I've also recently tried to get Opus 4.5 to move the Job system from Doom 3 BFG to the original codebase. Clean clone of dhewm3, pointed Opus to the BFG Job system codebase, and explained how it works. I have also fed it the Fabien Sanglard code review of the job system: https://fabiensanglard.net/doom3_bfg/threading.php
We are not sleeping on it, we are actually waiting for it to get actually useful. Sure, it can generate a full stack admin control panel in JS for my PostgreSQL tables, but is that really "not normal"? That's basic.
If you have any opensource examples of your codebase, prompt, and/or output, I would happily learn from it / give advice. I think we're all still figuring it out.
Also this SIMD translation wasn't just a single function - it was multiple functions across a whole region of the codebase dealing with video and frame capture, so pretty substantial.
That's a good way to say it, I totally identify.
I'm not talking about "write this function" but rather like implementing the whole feature by writing only English to the agent, over the course of numerous back-and-forth interactions and exhausting multiple 200K-token context windows.
For me personally, definitely at least 99% all of the Rust code I've committed at work since Opus 4.5 came out has been from an agent running that model. I'm reading lots of Rust code (that Opus generated) but I'm essentially no longer writing any of it. If dot-autocomplete (and LLM autocomplete) disappeared from IDE existence, I would not notice.
And also - doesn’t that make Zed (and other editors) pointless?
Nobody is one-shotting anything nontrivial in Zed's code base, with Opus 4.5 or any other model.
What about a future model? Literally nobody knows. Forecasts about AI capabilities have had horrendously low accuracy in both directions - e.g. most people underestimated what LLMs would be capable of today, and almost everyone who thought AI would at least be where it is today...instead overestimated and predicted we'd have AGI or even superintelligence by now. I see zero signs of that forecasting accuracy improving. In aggregate, we are atrocious at it.
The only safe bet is that hardware will be faster and cheaper (because the most reliable trend in the history of computing has been that hardware gets faster and cheaper), which will naturally affect the software running on it.
> And also - doesn’t that make Zed (and other editors) pointless?
It means there's now demand for supporting use cases that didn't exist until recently, which comes with the territory of building a product for technologists! :)
And yeah - if I had a crystal ball, I would be on my private island instead of hanging on HN :)
It takes more work than one-shot, but not a lot, and it pays dividends.
Are you worried about microsoft stealing your codebase from github?
And even with a narrower definition of stealing, Microsoft’s ability to share your code with US government agencies is a common and very legitimate worry in plenty of threat model scenarios.
The bigger a project gets the more context you generally need to understand any particular part. And by default Claude Code doesn't inject context, you need to use 3rd party integrations for that.
From what I've seen, CC has troubles with the latest Swift too, partially because of it being latest and partially because it's so convoluted nowadays.
But it's übercharged™ for C#
Because types are proofs and require global correctness, you can't just iterate, fix things locally, and wait until it breaks somewhere else that you also have to fix locally.
My rule of thumb is that if you can clearly describe exactly what you want to another engineer, then you can instruct the agent to do it too.
Why do we hold calculators to such high bars? Humans make calculation mistakes all the time.
Why do we hold banking software to such high bars? People forget where they put their change all the time.
Etc etc.
I've worked with Gemini Fast on the web to help design the VM ISA, then next steps will be to have some AI (maybe Gemini CLI - currently free) write an assembler, disassembler and interpreter for the ISA, and then the recursive descent compiler (written in C) too.
I already had Gemini 3.0 Fast write me a precedence climbing expression parser as a more efficient drop-in replacement for a recursive descent one, although I had it do that in C++ as a proof-of-concept since I don't know yet what C libraries I want to build and use (arena allocator, etc). This involved a lot of copy-paste between Gemini output and an online C++ dev environment (OnlineGDB), but that was not too bad, although Gemini CLI would have avoided that. Too bad that Gemini web only has "code interpreter" support for Python, not C and/or C++.
Using Gemini to help define the ISA was an interesting process. It had useful input in a "pair-design" process, working on various parts of the ISA, but then failed to bring all the ideas together into a single ISA document, repeatedly missing parts of what had been previously discussed until I gave up and did that manually. The default persona of Gemini seems not very well suited to this type of work flow where you want to direct what to do next, since it seems they've RL'd the heck out of it to want to suggest next step and ask questions rather than do what is asked and wait for further instruction. I eventually had to keep asking it to "please answer then stop", and interestingly quality of the "conversation" seemed to fall apart after that (perhaps because Gemini was now predicting/generating a more adversarial conversation than a collaborative one?).
I'm wondering/hoping that Gemini CLI might be better at working on documentation than Gemini web, since then the doc can be an actual file it is editing, and it can use it's edit tool for that, as opposed to hoping that Gemini web can assemble chunks of context (various parts of the ISA discussion) into a single document.
Another alternative (not recommended due to potential for "drift") is to use Gemini's Canvas capability where it is working on a document rather than a specification being spread out over Chat, but this document is fully regenerated for every update (unlike Claude's artifacts), so there is potential for it to summarize or drop sections of the document ("drift") rather than just making requested changes. Canvas also doesn't have Artifact's versioning to allow you to go back to undo unwanted drifts/changes.
What I’ve learned is that once you reach that point you’ve got to break that problem down into smaller pieces that the AI can work productively with.
If you’re about to start with Gemini-cli I recommend you look up https://github.com/github/spec-kit. It’s a project out of Microsoft/Github that encodes a rigorous spec-then-implement multi pass workflow. It gets the AI to produce specs, double check the specs for holes and ambiguity, plan out implementation, translate that into small tasks, then check them off as it goes. I don’t use spec-kit all the time, but it taught me that what explicit multi pass prompting can do when the context is held in files on disk, often markdown that I can go in and change as needed. I think it ask basically comes down to enforcing enough structure in the form of codified processes, self checks and/or tests for your code.
Pro tip, tell spec-kit to do TDD in your constitution and the tests will keep it on the rails as you progress. I suspect “vibe coding” can get a bad rap due to lack of testing. With AI coding I think test coverage gets more important.
I thought I'd revive it, but this time with Vulkan and no third-party dependencies (except for Vulkan)
4.5 Sonet, Opus and Gemini 3.5 flash has helped me write image decoders for dds, png jpg, exr, a wayland window implementation, macOS window implementation, etc.
I find that Gemini 3.5 flash is really good at understanding 3d in general while sonnet might be lacking a little.
All these sota models seem to understand my bespoke Lua framework and the right level of abstraction. For example at the low level you have the generated Vulkan bindings, then after that you have objects around Vulkan types, then finally a high level pipeline builder and whatnot which does not mention Vulkan anywhere.
However with a larger C# codebase at work, they really struggle. My theory is that there are too many files and abstractions so that they cannot understand where to begin looking.
Which makes sense. I'm sure there's lots of training data for React/HTML/CSS/etc. but much less with Swift, especially the newer versions.
Now, was the code quality any good? Beats me, I am not a swift developer. I did it partly as an experiment to see what Claude was currently capable of and partly because I wanted to test the feasibility of setting up a simple passive data logger for my truck.
I'm tempted to take another swing with Opus 4.5 for the science.
With some entirely novel work we're doing, it's actually a hindrance as it consistently tells us the approach isn't valid/won't work (it will) and then enters "absolutely right" loops when corrected.
I still believe those who rave about it are not writing anything I would consider "engineering". Or perhaps it's a skill issue and I'm using it wrong, but I haven't yet met someone I respect who tells me it's the future in the way those running AI-based companies tell me.
For some things you can fire up Claude and have it generate great code from scratch. But for bigger code bases and more complex architecture, you need to break it down ahead of time so it can just read about the architecture rather than analyze it every time.
> Create a CLAUDE.md for a c++ application that uses libraries x/y/z
[Then I edit it, adding general information about the architecture]
> Analyze the library in the xxx directory, and produce a xxx_architecture.md describing the major components and design
> /agent [let claude make the agent, but when it asks what you want it to do, explain that you want it to specialize in subsystem xxx, and refer to xxx_architecture.md
Then repeat until you have the major components covered. Then:
> Using the files named with architecture.md analyze the entire system and update CLAUDE.md to use refer to them and use the specialized agents.
Now, when you need to do something, put it in planning mode and say something like:
> There's a bug in the xxx part of the application, where when I do yyy, it does zzz, but it should do aaa. Analyze the problem and come up with a plan to fix it, and automated tests you can perform if possible.
Then, iterate on the plan with it if you need to, or just approve it.
One of the most important things you can do when dealing with something complex is let it come up with a test case so it can fix or implement something and then iterate until it's done. I had an image processing problem and I gave it some sample data, then it iterated (looking at the output image) until it fixed it. It spent at least an hour, but I didn't have to touch it while it worked.
It's also hard to steer the plan mode or have it remember some behavior that you want to enforce. It's much better to create a custom command with custom instructions that acts as the plan mode.
My system works like this:
/implement command acts as an orchestrator & plan mode, and it is instructed to launch predefined set of agents based on the problem and have them utilize specific skills. Every time /implement command is initiated, it has to create markdown file inside my own project, and then each subagent is also instructed to update the file when it finished working.
This way, orchestrator can spot that agent misbehaved, and reviewer agent can see what developer agent tried to do and why it was wrong.
Skills on the other hand are commands ON STEROIDS. They can be packaged with actual scripts and executables, the PEP723 Python style + uv is super useful.
I have one skill for example that uses Python+Treesitter to check the unit thest quality of a Go project. It does some AST magic to check the code for repetition, stupid things like sleeps and relative timestamps etc. A /command _can_ do it, but it's not as efficient, the scripts for the skill are specifically designed for LLM use and output the result in a hyper-compact form a human could never be arsed to read.
claude-code has a built in plugin that it can use to fetch its own docs! You don't have to ever touch anything yourself, it can add the features to itself, by itself.
The major benefit of agents is that it keeps context clean for the main job. So the agent might have a huge context working through some specific code, but the main process can do something to the effect of "Hey UI library agent, where do I need to put code to change the color of widget xyz", then the agent does all the thinking and can reply with "that's in file 123.js, line 200". The cleaner you keep the main context, the better it works.
At the same time, I think there's limitations to these tools and that I wont ever be able to achieve what I see others saying about 95% of code being AI written or leaving the AI to iterate for an hour. There's just too many weird little pitfalls in our work that the AI just cannot seem to avoid.
It's understandable, I've fallen victim to a few of them too, but I have the benefit of the ability to continuously learn/develop/extrapolate in a way that the LLM cannot. And with how little documentation exists for some of these things (MASQUE proxying for example) anytime the LLM encounters this code it throws a fit, and is unable to contribute meaningfully.
So thanks for your suggestions, it has made Claude better and clearly I was dragging my feet a little. At the very least, it's freed up a some more of my time to work on the complex things Claude can't do.
This is definitely not the case, and the reason anthropic doesnt make claude do this is because its quality degrades massively as you use up its context. So the solution is to let users manage the context themselves in order to minimize the amount that is "wasted" on prep work. Context windows have been increasing quite a bit so I suspect that by 2030 this will no longer be an issue for any but the largest codebases, but for now you need to be strategic.
Correct. In fact, this is the entire reason for the disconnect, where it seems like half the people here think LLMs are the best thing ever and the other half are confused about where the value is in these slop generators.
The key difference is (despite everyone calling themselves an SWE nowadays) there's a difference between a "programmer" and an "engineer". Looking at OP, exactly zero of his screenshotted apps are what I would consider "engineering". Literally everything in there has been done over and over to the death. Engineering is.. novel, for lack of a better word.
See also: https://www.seangoedecke.com/pure-and-impure-engineering/
Engineering is just problem solving, nobody judges structural engineers for designing structures with another Simpson Strong Tie/No.2 Pine 2x4 combo because that is just another easy (and therefore cheap) way to rapidly get to the desired state. If your client/company want to pay for art, that's great! Most just want the thing done fast and robustly.
I'm sure AI will get there, I also think it's not very good yet.
Tell that to the guys drawing up the world's 10 millionth cable suspension bridge
OP says, "BUT YOU DON’T KNOW HOW THE CODE WORKS.. No I don’t. I have a vague idea, but you are right - I do not know how the applications are actually assembled." This is not what I would call an engineer. Or a programmer. "Prompter", at best.
And yes, this is absolutely "lesser than", just like a middleman who subcontracts his work to Fiverr (and has no understanding of the actual work) is "lesser than" an actual developer.
I have a great time using Claude Code in Rust projects, so I know it's not about the language exactly.
My working model is is that since LLM are basically inference/correlation based, the more you deviate from the mainstream corpus of training data, the more confused LLM gets. Because LLM doesn't "understand" anything. But if it was trained on a lot of things kind of like the problem, it can match the patterns just fine, and it can generalize over a lot layers, including programming languages.
Also I've noticed that it can get confused about stupid stuff. E.g. I had two different things named kind of the same in two parts of the codebase, and it would constantly stumble on conflating them. Changing the name in the codebase immediately improved it.
So yeah, we've got another potentially powerful tool that requires understanding how it works under the hood to be useful. Kind of like git.
Even more basic failure mode. I told it to convert/copy a bit (1k LOC) of blocking code into a new module and convert to async. It just couldn't do a proper 1:1 logical _copy_. But when I manually `cp <src> <dst>` the file and then told it to convert that to async and fix issues, it did it 100% correct. Because fundamentally it's just non-deterministic pattern generator.
The Claude models are technically multi-modal, but IME the vision side of the equation is really lacking. As a result, Claude is quite good at reasoning about logic, and it can build e.g. simpler web pages where the underlying html structure is enough to work with, but it’s much worse at tasks that inherently require seeing.
I have not had this experience at all. It often doesn't get it right on the first pass, yes, but the advantage with Rust vibecoding is that if you give it a rule to "Always run cargo check before you think you're finished" then it will go back and fix whatever it missed on the first pass. What I find particularly valuable is that the compiler forces it to handle all cases like match arms or errors. I find that it often misses edge cases when writing typescript, and I believe that the relative leniency of the typescript compiler is why.
In a similar vein, it is quite good at writing macros (or at least, quite good given how difficult this otherwise is). You often have to cajole it into not hardcoding features into the macro, but since macros resolve at compile time they're quite well-suited for an LLM workflow as most potential bugs will be apparent before the user needs to test. I also think that the biggest hurdle of writing macros to humans is the cryptic compiler errors, but I can imagine that since LLMs have a lot of information about compilers and syntax parsing in their training corpus, they have an easier time with this than the median programmer. I'm sure an actual compiler engineer would be far better than the LLM, but I am not that guy (nor can I afford one) so I'm quite happy to use LLMs here.
For context, I am purely a webdev. I can't speak for how well LLMs fare at anything other than writing SQL, hooking up to REST APIs, React frontend, and macros. With the exception of macros, these are all problems that have been solved a million times thus are more boilerplate than novelty, so I think it is entirely plausible that they're very poor for different domains of programming despite my experiences with them.
for example, one of our public repos works with rust compiler artifacts and cache restoration (https://github.com/attunehq/hurry); if you look at the history you can see it do some pretty surprisingly complex (and well made, for an LLM) changes. its code isn't necessarily what i would always write, or the best way to solve the problem, but it's usually perfectly serviceable if you give it enough context and guidance.
> we have another Claude Code agent that does a full PR review, following a detailed markdown checklist we’ve written for it.
I shortened it for anyone else that might need it
----
Software engineers are sleeping on Claude Code agents. By teaching it your conventions, you can automate your entire workflow:
Custom Skills: Generates code matching your UI library and API patterns.
Quality Ops: Automates ESLint, doc syncing, and E2E coverage audits.
Agentic Reviews: Performs deep PR checks against custom checklists.
Smart Triage: Pre-analyzes tickets to give devs a head start.
Check out the showcase repo to see these patterns in action.
Especially if you're in a place where a lot of time was spent previously revising PRs for best practices, etc, even for human-submitted code, then having the LLM do that for you that saves a bunch of time. Most humans are bad at following those super-well.
There's a lot of stuff where I'm pretty sure I'm up to at least 2x speed now. And for things like making CLI tools or bash scripts, 10x-20x. But in terms of "the overall output of my day job in total", probably more like 1.5x.
But I think we will need a couple major leaps in tooling - probably deterministic tooling, not LLM tooling - before anyone could responsibly ship code nobody has ever read in situations with millions of dollars on the line (which is different from vibe-coding something that ends up making millions - that's a low-risk-high-reward situation, where big bets on doing things fast make sense. if you're already making millions, dramatic changes like that can become high-risk-low-reward very quickly. In those companies, "I know that only touching these files is 99.99% likely to be completely safe for security-critical functionality" and similar "obvious" intuition makes up for the lack of ability to exhaustively test software in a practical way (even with fuzzers and things), and "i didn't even look at the code" is conceding responsibility to a dangerous degree there.)
claude can call ssh and do system admin tasks. It works amazingly well. I have 3 VM's, which depends on each other (proxmox with openwrt, adguard, unbound), and claude can prove to me that my dns chains works perfectly, my firewalls are perfect etc as claude can ssh into each. Setting up services, diagnosing issues, auditing configs... you name it. Just awesome.
claude can call other sh scripts on the machine, so over time, you can create a bunch of scripts that lets claude one shot certain tasks that would normally eat tokens. It works great. One script per intention - don't have a script do more than one thing.
claude can call the compiler, run the debug executable and read the debug logs.. in real time. So claude can read my android apps debug stream via adb.. or my C# debug console because claude calls the compiler, not me. Just ask it to do it and it will diagnose stuff really quickly.
It can also analyze your db tables (give it readonly sql access), look at the application code and queries, and diagnose performance issues.
The opportunities are endless here. People need to wake up to this.
Claude set up a Raspberry Pi with a display and conference audio device for me to use as an Alexa replacement tied to Home Assistant.
I gave it an ssh key and gave it root.
Then I told it what I wanted, and it did. It asked for me to confirm certain things, like what I could see on screen, whether I could hear the TTS etc. (it was a bit of a surprise when it was suddenly talking to me while I was minding my own business).
It configured everything, while keeping a meticulous log that I can point it at if I want to set up another device, and eventually turn into a runbook if I need to.
I'm not saying there aren't use cases for agents, just that it's normal that most software engineers are sleeping on it.
I have Django project with 50 kLOC and it is pretty capable of understanding the architecture, style of coding, naming of variables, functions etc. Sometimes it excels on tasks like "replicate this non-trivial functionality for this other model and update the UI appropriately" and leaves me stunned. Sometimes it solves for me tedious and labourous "replace this markdown editor with something modern, allowing fullscreen edits of content" and does annoying mistake that only visual control shows and is not capable to fix it after 5 prompts. I feel as I am becoming tester more than a developer and I do not like the shift. Especially when I do not like to tell someone he did an obvious mistake and should fix it - it seems I do not care if it is human or AI, I just do not like incompetence I guess.
Yesterday I had to add some parameters to very simple Falcon project and found out it has not been updated for several months and won't build due to some pip issues with pymssql. OK, this is really marginal sub-project so I said - let's migrate it to uv and let's not get hands dirty and let the Claude do it. He did splendidly but in the Dockerfile he missed the "COPY server.py /data/" while I asked him to change the path... Build failed, I updated the path myself and moved on.
And then you listen to very smart guys like Karpathy who rave about Tab, Tab, Tab, while not understanding the language or anything about the code they write. Am I getting this wrong?
I am really far far away from letting agents touch my infrastructure via SSH, access managed databases with full access privileges etc. and dread the day one of my silly customers asks me to give their agent permission to managed services. One might say the liability should then be shifted, but at the end of the day, humans will have to deal with the damage done.
My customer who uses all the codebase I am mentioning here asked me, if there is a way to provide "some AI" with item GTINs and let it generate photos, descriptions, etc. including metadata they handcrafted and extracted for years from various sources. While it looks like nice idea and for them the possibility of decreasing the staff count, I caught the feeling they do not care about the data quality anymore or do not understand the problems the are brining upon them due to errors nobody will catch until it is too late.
TL;DR: I am using Opus 4.5, it helps a lot, I have to keep being (very) cautious. Wake up call 2026? Rather like waking up from hallucination.
take my downvote as hard as you can. this sort of thing is awfully off-putting.
The tech industry just went through an insane hiring craze and is now thinning out. This will help to separate the chaff from the wheat.
I don't know why any company would want to hire "tech" people who are terrified of tech and completely obstinate when it comes to utilizing it. All the people I see downplaying it take a half-assed approach at using it then disparage it when it's not completely perfect.
I started tinkering with LLMs in 2022. First use case, speak in natural english to the llm, give it a json structure, have it decipher the natural language and fill in that json structure (vacation planning app, so you talk to it about where/how you want to vacation and it creates the structured data in the app). Sometimes I'd use it for minor coding fixes (copy and paste a block into chatgpt, fix errors or maybe just ideation). This was all personal project stuff.
At my job we got LLM access in mid/late 2023. Not crazy useful, but still was helpful. We got claude code in 2024. These days I only have an IDE open so I can make quick changes (like bumping up a config parameter, changing a config bool, etc.). I almost write ZERO code now. I usually have 3+ claude code sessions open.
On my personal projects I'm using Gemini + codex primarily (since I have a google account and chatgpt $20/month account). When I get throttled on those I go to claude and pay per token. I'll often rip through new features, projects, ideas with one agent, then I have another agent come through and clean things up, look for code smells, etc. I don't allow the agents to have full unfettered control, but I'd say 70%+ of the time I just blindly accept their changes. If there are problems I can catch them on the MR/PR.
I agree about the low hanging fruit and I'm constantly shocked at the sheer amount of FUD around LLMs. I want to generalize, like I feel like it's just the mid/jr level devs that speak poorly about it, but there's definitely senior/staff level people I see (rarely, mind you) that also don't like LLMs.
I do feel like the online sentiment is slowly starting to change though. One thing I've noticed a lot of is that when it's an anonymous post it's more likely to downplay LLMs. But if I go on linkedin and look at actual good engineers I see them praising LLMs. Someone speaking about how powerful the LLMs are - working on sophisticated projects at startups or FAANG. Someone with FUD when it comes to LLM - web dev out of Alabama.
I could go on and on but I'm just ranting/venting a little. I guess I can end this by saying that in my professional/personal life 9/10 of the top level best engineers I know are jumping on LLMs any chance they get. Only 1/10 talks about AI slop or bullshit like that.
The "sudden switch" since Opus 4.5 when many were saying just a few months ago "I enjoy actual coding" but now are praising LLM's isn't a one off occurrence. I do think underneath it is somewhat motivated by fear; not for the job however but for relevance. i.e. its in being relevant to discussions, tech talks, new opportunities, etc.
I agree with this, but I haven't needed to use any advanced features to get good results. I think the simple approach gets you most of the benefits. Broadly, I just have markdown files in the repo written for a human dev audience that the agent can also use.
Basically:
- README.md with a quick start section for devs, descriptions of all build targets and tests, etc. Normal stuff.
- AGENTS.md (only file that's not written for people specifically) that just describes the overall directory structure and has a short step of instructions for the agent: (1) Always read the readme before you start. (2) Always read the relevant design docs before you start. (3) Always run the linter, a build, and tests whenever you make code changes.
- docs/*.md that contain design docs, architecture docs, and user stories, just text. It's important to have these resources anyway, agent or no.
As with human devs, the better the docs/requirements the better the results.
It's a tiny step further, and sub-agents provide a massive benefit the moment you're ready to trust the model even a little bit (relax permissions to not have it prompt you for every little thing; review before committing rather than on every file edit) because they limit what goes into the top level context, and can let the model work unassisted for far longer. I now regularly have it run for hours at a time without stopping.
Running and acting on output from the linter is absolutely an example of that which matters even for much shorter runs.
There's no reason to have all the lint output "polluting" the top level context, nor to have the steps the agent needs to take to fix linter issues that can't be auto-fixed by the linter itself. The top level agent should only need to care about whether the linter run passed or failed (and should know it needs to re-run and possibly investigate if it fails).
Just type /agents, select "Create new agent" and describe a task you often do, and then forget about it (or ask Claude to make changes to it for you)
I was trying to get across the point that today you can get a lot of benefit from minimal setup, even one that's vendor-agnostic. (The steps I outlined work for Codex out of the box, too.)
You're right to point out that the more you refine things, the more you'll get out of the tools. It used to be that you had to do a lot of refinements to start getting good results at all. Now, you can get a lot out of even a basic setup like I outlined, which is great for people who are new users -- or people who tried it before and weren't that impressed but are now giving it another try.
There are times those extra layers are worth it but it seems LLMs have a bias to add them prematurely and overcomplicate things. You then end up with extra complexity you didn't need.
Once you’ve got Claude Code set up, you can point it at your codebase, have it learn your conventions, pull in best practices, and refine everything until it’s basically operating like a super-powered teammate. The real unlock is building a solid set of reusable “skills” plus a few agents for the stuff you do all the time.
For example, we have a custom UI library, and Claude Code has a skill that explains exactly how to use it. Same for how we write Storybooks, how we structure APIs, and basically how we want everything done in our repo. So when it generates code, it already matches our patterns and standards out of the box.
We also had Claude Code create a bunch of ESLint automation, including custom ESLint rules and lint checks that catch and auto-handle a lot of stuff before it even hits review.
Then we take it further: we have a deep code review agent Claude Code runs after changes are made. And when a PR goes up, we have another Claude Code agent that does a full PR review, following a detailed markdown checklist we’ve written for it.
On top of that, we’ve got like five other Claude Code GitHub workflow agents that run on a schedule. One of them reads all commits from the last month and makes sure docs are still aligned. Another checks for gaps in end-to-end coverage. Stuff like that. A ton of maintenance and quality work is just… automated. It runs ridiculously smoothly.
We even use Claude Code for ticket triage. It reads the ticket, digs into the codebase, and leaves a comment with what it thinks should be done. So when an engineer picks it up, they’re basically starting halfway through already.
There is so much low-hanging fruit here that it honestly blows my mind people aren’t all over it. 2026 is going to be a wake-up call.
(used voice to text then had claude reword, I am lazy and not gonna hand write it all for yall sorry!)
Edit: made an example repo for ya
https://github.com/ChrisWiles/claude-code-showcase