581–590 of 1001 posts
I hope AI leads to a Cambrian explosion of software people running their own businesses, given the force multiplier it affords. On the other hand, the jaded part of me feels that AI may lead to a consolidation into a very small set of monopolies. We'll see.
I don't understand the stance that AI currently is able to automate away non-trivial coding tasks. I've tried this consistently since GPT 3.5 came out, with every single SOTA model up to GPT 5.1 Codex Max and Opus 4.5. Every single time, I get something that works, yes, but then when I start self-reviewing the code, preparing to submit it to coworkers, I end up rewriting about 70% of the thing. So many important deta…
For personal projects on the other hand, it has expedited me what? 10x, 30x? It's not measurable. My output has been so much more than what would have been possible earlier, that there is no benchmark because these level of projects would not have been getting completed in the first place.
Back to using at work: I think it's a skill issue. Both on my end and yours. We haven't found a way to encode our domain knowledge into AI and transcend into orchestrators of that AI.
> How do I feel, about all the code I wrote that was ingested by LLMs? I feel great to be part of that, because I see this as a continuation of what I tried to do all my life: democratizing code, systems, knowledge. LLMs are going to help us to write better software, faster, and will allow small teams to have a chance to compete with bigger companies. Every now and then I post the same exact comment here on HN, where…
how AI speeds me up:
- no longer have to remember how to set up unit test boilerplate in each of the 6ish programming languages i commonly use
- can often vaguely gesture at an existing pattern and have AI "copy-paste" it into new code. "do that read-through cache pattern like you see there and there but do it for this table and this proto msg type."
- can quickly answer questions like "does anyone in the code seem to build this string manually instead of using the library/helper method for it"
- can quickly generate code like "all I want is a gosh dang PKCS-formatted key, why is that so hard for this library" which the docs did not provide
which is really cool. it absolutely speeds things up by 10-100x in some scenarios. a lot of the sucky parts of programming are being mired down in these kinds of messes.
how AI slows me down:
- have to explain to jr dev why, even though it has unit tests, the AI-generated bespoke mutex async cache is not going into our production codebase
- have to explain to PM why I cannot let them vibe code new features into the hot path of our prod services when they are not on-call to be forced to clean it up when it explodes at 3am
- have to explain to senior dev who should REALLY know better why you cannot _just_ ask someone to review a 2000 LOC PR
- have to explain to CEO in tremendous itemized, evidenced detail why [big project in eye of sauron] did not go noticeably faster than it did 2 years ago even though the team was hand-picked to be full of people he knew would use AI as much as he wanted them to.
- have to explain to CEO why I really wish he would stop playing with AI and bothering the crap out of the engineers and go back to actually doing whatever it is the CEO gets paid 10-100x what a software engineer salary to do. [actually still trying to figure this one out without getting fired.]
I'm as interested in AI use as anyone can be, when I have to put up with sycophantic "believers" who really wish they could replace me entirely with the chatbot.
Also, this shit is expensive and still being sold at a loss. I signed up for Amp and blew through my $10 of signup credit getting very little done. I'm certainly not paying my own money for that.
> But what was the fire inside you, when you coded till night to see your project working? It was building. I feel like this is not the same for everyone. For some people, the "fire" is literally about "I control a computer", for others "I'm solving a problem for others", and yet for others "I made something that made others smile/cry/feel emotions" and so on. I think there is a section of programmer who actually do…
I don't think this is really it for many people (maybe any); after all, you can do all of that when writing a text message rather than a piece of code.
But it inches closer to what I think is the "right answer" for this type of software developer. There are aspects of software development that are very much like other forms of writing (e.g., prose or poetry).
Like other writing, writing code can constitute self-expression in an inherently satisfying way, and it can also offer the satisfaction of finding "the perfect phrase". LLMs more or less eliminate both sources of pleasure, either by eliminating the act of writing itself (that is, choosing and refining the words) or through their bland, generic, tasteless style.
There are other ways that LLMs can disconnect the people using them from what is joyful about writing code, not least of all because LLMs can be used in a lot of different ways. (Using them as search tools or otherwise consulting them rather than having them commit code to simply be either accepted/rejected "solves" the specific problems I just mentioned, for instance.)
There is something magical about speaking motion into existence, which is part of what has made programming feel special to me, ever since I was a kid. In a way, prompting an LLM to generate working code preserves that and I can imagine how, for some, it even seems to magnify the magic. But there is also a sense of essential mastery involved in the wonderful way code brings ideas to life. That mastery involves not just "understanding" things in the cursory way involved in visually scanning someone else's code and thinking "looks good to me", but intimately knowing how the words and abstractions and effects all "line up" and relate to each other (and hopefully also with the project's requirements). That feeling of mastery is itself one of the joys of writing code.
Without that mastery, you also lose one of the second-order joys of writing code that many here have already mentioned in these comments: flow. Delegation means fumbling in a way that working in your own context just doesn't. :-\
> But what was the fire inside you, when you coded till night to see your project working? It was building. I feel like this is not the same for everyone. For some people, the "fire" is literally about "I control a computer", for others "I'm solving a problem for others", and yet for others "I made something that made others smile/cry/feel emotions" and so on. I think there is a section of programmer who actually do…
> I think there is a section of programmer who actually do like the actual typing of letters, numbers and special characters into a computer, and for them, I understand LLMs remove the fun part. I've "vibe coded" a ton of stuff and so I'm pretty bullish on LLMs, but I don't see a world where "coding by hand" isn't still required for at least some subset of software. I don't know what that subset will be, but I'm conv…
I don't understand the stance that AI currently is able to automate away non-trivial coding tasks. I've tried this consistently since GPT 3.5 came out, with every single SOTA model up to GPT 5.1 Codex Max and Opus 4.5. Every single time, I get something that works, yes, but then when I start self-reviewing the code, preparing to submit it to coworkers, I end up rewriting about 70% of the thing. So many important deta…
On the subpar code, would the code work, albeit suboptimally? I think part of the problem a lot of senior devs are having is that they see what they do as an artisanal craft. The rest of the world just sees the code as a means to an end. I don't care how elegantly my toaster was crafted as long as it toasts the bread and doesn't break.
- it becomes brittle and rigid (can't change it, can't add to it)
- it becomes buggy and impossible to fix one bug without creating another
- it becomes harder to tell what it's doing
- plus it can be inefficient / slow / insecure, etc.
The problem with your analogy is that toasters are quite simple. The better example would be your computer, and if you want your computer to just run your programs and not break, then these things matter.
I don't understand the stance that AI currently is able to automate away non-trivial coding tasks. I've tried this consistently since GPT 3.5 came out, with every single SOTA model up to GPT 5.1 Codex Max and Opus 4.5. Every single time, I get something that works, yes, but then when I start self-reviewing the code, preparing to submit it to coworkers, I end up rewriting about 70% of the thing. So many important deta…
I don't understand the stance that AI currently is able to automate away non-trivial coding tasks. I've tried this consistently since GPT 3.5 came out, with every single SOTA model up to GPT 5.1 Codex Max and Opus 4.5. Every single time, I get something that works, yes, but then when I start self-reviewing the code, preparing to submit it to coworkers, I end up rewriting about 70% of the thing. So many important deta…
> Other people are just less picky than I am I think this is part of it. When coding style has been established among a team, or within an app, there are a lot of extra hoops to jump through, just to get it to look The Right Way, with no detectable benefit to the user. If you put those choices aside and simply say: does it accomplish the goal per the spec (and is safe and scalable[0]), then you can get away with a lo…
It feels like you're diminishing the parent commenter's views, reducing it to the perspective of style. Their comment didn't mention style.
I don't understand the stance that AI currently is able to automate away non-trivial coding tasks. I've tried this consistently since GPT 3.5 came out, with every single SOTA model up to GPT 5.1 Codex Max and Opus 4.5. Every single time, I get something that works, yes, but then when I start self-reviewing the code, preparing to submit it to coworkers, I end up rewriting about 70% of the thing. So many important deta…
Instead of rewriting yourself have you tried telling the agent what it did wrong and do the rewrite with it? Then at the end of the session ask it to extract a set of rules that would have helped to get it right the first time. Save that in AGENTS.md. If you and your team do this a few times it can lead to only having to rewrite 5% of the code instead of 70%.
I have, it becomes a race to the bottom.