A Hacker News thread titled “How do we stop vibe coding?” exploded this week with over 70 comments — and the opinions were all over the map. Some developers called vibe coding a symptom of laziness. Others said it is the best thing to happen to software in decades. A few described real workflows that would surprise both camps.
If you build software with AI tools, or you are deciding whether to start, the raw developer debate is more useful than any product announcement. Here is what people actually said, what surprised us, and what it means if you write code for a living.
First: what developers actually mean by “vibe coding”
Before you can evaluate anyone’s opinion, you have to understand that developers cannot agree on what vibe coding even means. The term was coined by Andrej Karpathy, who described it as fully giving in to the vibes, embracing exponentials, and forgetting that the code even exists. On Hacker News, that definition became the center of a fight.
One developer insisted that vibe coding means you turn your brain off entirely. Another shot back that they had never been more informed about their own code, pipelines, and stack than since they started using AI agents. “It’s not some intelligence dementor,” one commenter wrote. “If your brain turned off while you were vibe coding, that’s honestly just a you problem.”
A third commenter tried to draw a line: using an LLM to help you write code while keeping your brain on is not vibe coding — it is AI-assisted development. Vibe coding, in their view, means you genuinely do not care what the agent produces as long as it runs. That distinction matters because the two activities have completely different risk profiles.
The skills atrophy argument — and why it is more nuanced than you think
The most emotionally charged comments were about whether AI coding tools make developers dumber. One developer cited studies showing that skills atrophy is real. “If your brain does not turn off while you are vibe-coding, keep it up and it soon will,” they warned.
Another developer compared it to using a calculator. When you stop doing mental arithmetic, you lose the ability. But that does not mean calculators are bad — it means you have to choose when to practice mental math versus when to use the tool.
The most interesting take came from a developer who described what they called “rawdog Thursdays”: every Thursday, they write all their code without any AI assistant, not even inline suggestions. “Allows them to maintain their craft and keep their codebase human maintainable, while still benefiting from the tool the rest of the week,” they wrote. It is a practical compromise between the purists who refuse AI entirely and the full-automation crowd.
A senior developer added that the atrophy concern is real but overstated for experienced engineers. “I have established many skills that have atrophied which I have restored in need and/or desire over a span of decades,” they wrote. Younger developers, they cautioned, should be more careful — they have not yet built the deep foundations that let you recover later.
The delegation insight: good leads are good at this
One of the most surprising comments reframed the entire debate. A commenter noted that the developers struggling most with AI coding are “the ones that never got to grips with human delegation either.” In their experience, it is the experienced leads and senior engineers — the ones who predate the Agile “everyone is equal” era — who get the most real mileage out of AI agents.
That connection between delegation skill and AI skill is not obvious, but it makes sense. Both require you to break a big task into small, well-scoped pieces. Both require clear acceptance criteria. Both require a trust-but-verify approach where you review output without micromanaging every step.
Another developer pushed back on the analogy, pointing out that when you delegate to humans, you have a baseline level of trust in their competence. With AI agents, you cannot trust anything — they might produce brilliant code or confidently generate a function that crashes in production. “You still have to apply engineering thinking to human and LLM organization,” they wrote.
The burnout paradox: AI coding can be MORE exhausting
Several developers described a counterintuitive experience: AI-assisted coding is more tiring than doing it by hand. One commenter agreed that it is “actually more exhausting in some ways.” The reason became clear in the follow-up discussion.
When you write code manually, the cognitive load is steady and predictable. You know what you know, and you work within that boundary. When you use an AI agent, you are simultaneously directing, reviewing, debugging misunderstandings, and catching subtle errors in code you did not write. The mental overhead of managing an agent on complex tasks can exceed the overhead of just writing the code yourself.
One developer described the common pattern: spending a long time crafting an elaborate prompt, then arguing with the agent for 10 to 20 turns, getting a passable plan, accepting it, then arguing for another 10 to 20 turns while it implements. “There’s almost always a misunderstanding within the smaller tasks anyway, which I have to test to find out about, then ask the agent to fix,” they wrote.
But other developers described completely different experiences. “Both myself and the AI correct and remove each other’s gaps in understanding. There is never what I would call arguing,” one commenter reported. The difference seems to come down to how you scope the work. Developers who break problems into tiny, verifiable pieces report smooth experiences. Developers who give large, ambiguous tasks report frustration.
The economics argument: bespoke coding may not survive
A developer named thenatureboy laid out the uncomfortable math: “Economics will dictate that working bespoke on coding will simply not be profitable anymore.” They were not celebrating this — they were observing it.
The counterarguments were immediate. One developer pointed out that if pure economics ruled, there would be no good San Francisco tech jobs — all coding would happen in cheaper labor markets. Cultural assumptions, relationships, and trust layers protect developer compensation in ways that pure economics do not predict.
Another commenter drew a comparison to photography. Stock photography is now everywhere, but that did not kill custom photo shoots. “Most of the commercial stuff — copy/ads with generic guy in suit picture — will be AI that is good enough,” one developer predicted. The bespoke, complex work will remain human, but the volume of that work may shrink.
A software engineer who has been coding for 15 years put it more personally: “If anything is leading me to burnout, it’s this bullshit hype cycle.” They described how coding has been one of the greatest jobs they could imagine — earning money for solving puzzles. “With vibe coding that joy of solving a complex problem is replaced with the tedium of prompting and reviewing,” another commenter agreed.
The sandbox security angle nobody is talking about
While the vibe coding debate raged, a separate Hacker News thread about “Sandbox Escape Vulnerabilities Across 4 Coding Agent Vendors” quietly gathered attention. The key insight: “Sandboxed” does not necessarily mean “safe.”
One developer pointed out that the most concerning part was not the escape itself, but “all the trusted software that unknowingly helps it.” AI agents with access to your terminal, file system, and Git repository are a larger attack surface than most developers realize. This is particularly relevant for vibe coding, where developers are less likely to scrutinize what the agent is actually doing under the hood.
What this means for you
If you write code — whether full-time, part-time, or as a hobby — the HN debate surfaces three practical lessons:
First, the developers who thrive with AI coding tools are the ones who already knew how to break problems into small pieces. If you struggle with agent-based coding, the fix might not be a better model or a better prompt. It might be better decomposition of the problem itself.
Second, the skill atrophy concern is real but manageable. The “rawdog one day a week” approach is simple and effective. If you are a junior developer, be especially cautious about letting AI do all your learning for you. The developers who can recover from atrophy are the ones who built deep foundations first.
Third, AI coding is not one thing. The developers reporting exhaustion and the developers reporting breakthroughs are both telling the truth — about their own workflows. The variable is not the model. It is the process. Developers who maintain review discipline, test coverage, and human oversight get the productivity boost. Developers who hand everything to the agent and hope for the best get the burnout.
The vibe coding debate is not going away. As one developer put it: “The thing is out of the bottle, can’t stop it.” But the terms of engagement are still up to you.
Building with AI coding tools? aiFiesta gives you access to 9+ premium models for $12/mo instead of $20/mo each — more models to test, less money to spend.