this post was submitted on 28 Jun 2026
72 points (92.9% liked)
Programming
27489 readers
398 users here now
Welcome to the main community in programming.dev! Feel free to post anything relating to programming here!
Cross posting is strongly encouraged in the instance. If you feel your post or another person's post makes sense in another community cross post into it.
Hope you enjoy the instance!
Rules
Rules
- Follow the programming.dev instance rules
- Keep content related to programming in some way
- If you're posting long videos try to add in some form of tldr for those who don't want to watch videos
Wormhole
Follow the wormhole through a path of communities !webdev@programming.dev
founded 3 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
As a senior developer I have serious doubts about the whole thing. Yes, I don't do tedious typing anymore, now I do extremely tedious code review all day, my least favorite part of the job. And I have to be very vigilant because the AI is an idiot more often than not. Then when I finally publish my own code it's time to go review my colleagues' ai code and figure out what they missed in their review.
I don't feel much of a productivity multiplier. I'm not saying we won't get there, but this current iteration ain't it.
you are a senior developer, start to transform your view in a senior architect. With AI there's no more need of developers. An architect ius needed, and if it haas a solid dev backround as yours projects will change view. Technology is always changing and it's hard to stay at pace. But if you look from an higher perspective your project your experience will only help the AI to do the Job for you
lol no. this is what a mediocre dev thinks.
In your mental model, you can only become an architect if your artificial subordinate does its job properly. Unfortunately, it looks like the subordinate is an idiot, so your analogy just doesn't work.
I think in your mental model lies the very popular misconception that humans are any good at coding, and that architects were able to do their jobs because they were sitting on top of competent operators.
I'd argue that this is wholly untrue. In fact, for 30 years the software development field has produced mountains of sociology and processes designed to coerce good software out of idiots writing arbitrary code. Idiot subordinates is the baseline here, not an anomaly introduced by AI.
I'd even go further and say that current gen AI is marginally better than the average developer so as an architect you're still herding cats but the cats are marginally less crazy than they were, say, 10 years ago. The methods are roughly the same : deep roadmaps, shallow sprints, frequent iterations and constant supervision. It's not ideal but it has produced all the software known to man, including critical life-or-death stuff.
At least, human agents can take responsibility for their actions, and they learn. When they fail to do so, they get fired.
You're right, it's not only that the subordinate is an idiot. It's also a pathological liar that never learns and can't take the blame for their fuckups.
If a subordinate persistently sneaks in bad code despite being told not to, this is grounds for dismissal as far as I'm concerned.
It looks like you're attributing human intentions to a machine, that can't be healthy. You're also dismissing how humans routinely lie, deflect blame (especially when their job is at stake), or fail to learn for a myriad of valid reasons.
Agents don't "sneak in" bad code, or "lie". They get shit wrong sometimes, or they get lazy at the tail end of a long task, which are not categorically new things in software development. Even excellent engineers do it. If your only response to that is to say "yeah but i can fire them" then you are a bad lead. Just some corpo mid-manager eager to push blame and consequences at your underlings instead of creating a system that allows them to perform within their individual limitations.
You're right that they don't learn in that the model doesn't re-weight itself based on your feedback but that one is so trivially solved that it's not even a subject. Haven't you ever worked with brilliant stoner/creative types ? They forget shit all the time so you learn to write everything down and devise stratagems so the relevant information they keep forgetting is pushed to the front of their mind regularly. Well, i'd do that, but you'd just fire them on the spot i guess.
My point is : on the one hand you're overestimating the failure modes of AI, and on the other hand you're dismissing the failure modes of humans. With the added simplistic authoritarian thinking "i'd fire them anyways if they don't behave".
Being a good lead means knowing when someone's not going to cut it, and when to let them go to protect the rest of the team. My boss recently failed to do so, and 3 people got laid off as a result because the money spent on the rotten apple was used compensating for their lackluster output and could not be used to pay the good employees' salaries.
Additionally, it also means seeing through your colleagues' bullshit (blame deflection, deception) and reacting accordingly. And even when you see through the bullshit, you sometimes need to compromise and learn to tolerate it because the tradeoff is worthwile. Sometimes.
Thanks for judging my performance as a lead based on fuck all. Now, moving on the point at hand.
You're spot on when saying that LLMs are not human beings and don't have intentions. The truth is, I don't actually care. The tech is advertized as a replacement for human programmers, and I judge on its actions in that context.
Moving back to your point - you're right that it doesn't lie. It's an analogy. However, it gaslights like a pathological liar and does the things it's been specifically told to not do, and does so repeatedly. If a human acted similarly, it'd be gone without delay, even when omitting the quality issues it has which humans also have.
This is a weird outcome and contrary to the experiences of most users. The tendency that is commonly observed is that, on the contrary, agents don't really "hold opinions" and flip-flop the moment you start challenging them. I'd be curious to learn what model and setup gave you an agent that defends its hallucinations because that is definitely an outlier in a world of sycophantic models that have been RLHF'd to hell and back.
In my experience, this is a tell that you are holding it very wrong. You can shit on models for having no creativity or losing track of long horizon tasks but following instructions is the one thing they do well. Again, stupid model or stupid harness, or you're just not holding the pointy end in the right direction.
US tech marketing is done by assholes targeting morons. It is probably safer to never, ever take their statements at face value.
There are 2 categories of people who are dead wrong about agentic coding : rabid vibe-coders who think it 10x their productivity (it doesn't, it just exacerbates their worst tendencies), and people like you who think it's good for nothing. Both categories are just vibing based on a gut feeling. But in the middle there is a whole spectrum of actually interesting experimentation, if you are ready to fist fight your personal assumptions and be fact-based about it.
It's not weird. The good side of the LLM craze is that its failures are documented at length all over the Internet. It's in every tech blog out there, even in those run by users of agents. There isn't a day without a LLM article on the front page of HN and Lobster.rs.
I don't actually hate the tech itself, and, like most, am pretty impressed at what it can do at the moment - the way I'm impressed by what the useless Boston Dynamics robots can do.
I oppose LLM usage for practical, very rational reasons. The most important aspect is an ethical one: the planet is burning, I don't want to make things worse by using something that barely even works.
There is another, very important aspect: I can't actually run it locally, which means relying on it means depending on external suppliers from an enemy country that may cut the pipes without prior notice, as it tried to do very recently.
Additionally, the costs are very high, increasing and enshittification has barely even started. Once monopolies are established, I don't want to have to pay soaring cartel prices for a shitty service I could have avoided depending upon to begin with.
There's still a broad range of domains that are fundamentally incompatible with usage of LLMs because of reliability and/or confidentiality reasons, and I want to remain employable as a software engineer and stand out from the legion of CRUD developers whose skills will inevitably atrophy as LLM adoption increases.
I do see how I could use them sporadically, but cf. reason #1. Using LLMs is immoral.
And I do know how to do my job. I don't have a velocity issue. As they say, if it ain't broke, don't fix it.