unknown

Stack Overflow's 2025 Developer Survey reveals a paradox: 84% of developers now use AI tools (up from 76% last year), but trust has plummeted to 60% from over 70% in previous years. Only 33% trust AI accuracy, with just 3% having high trust. The main frustration is the "almost right" problem where 66% say AI gives solutions that appear correct but contain subtle errors. Developers embrace AI for routine tasks like searching answers (54%) but reject it for critical functions like deployment (76% won't use) and project planning (69% refuse). ChatGPT leads adoption at 82%, followed by GitHub Copilot at 68%. Most developers (72%) reject "vibe coding" and still prefer human oversight for quality control and complex decision-making.

Stack Overflow Survey Reveals Developers Are Using AI Tools More but Trust Them Less Than Ever

kt123123123

As a long-time senior dev &amp; CTO, we use AI sparingly. We have an unwritten rule that AI should represent no more than 10% of the code you ship. Any more than that, and:
<ol>
<li>You risk potentially not understanding what AI has written if it’s beyond your own capabilities as a developer.</li>
<li>You now depend on AI to address any bugs that arise as a result of the code that it wrote—that you “vibe coded.”</li>
</ol>
I will admit that we have, in a couple of instances, leaned heavily on AI to write stuff that was algorithmically complex, or to tackle tasks in one hour that would’ve taken a senior dev two weeks of research plus dev time. In those rare instances, when AI is having a “good day,” it will solve a complex issue in 1-2 rounds of prompting, and lets us focus on more important matters. Of course, a human being is always heavily QA’ing whatever AI wrote.
My recommendation is to use AI as a pair programmer, not a replacement for a coder. When you sit down with it and rubber-duck as you build something, it can actually be helpful in providing insights, teaching you about new things you may not have known about, and catching potential mistakes. But again, YMMV as Anthropic/OpenAI are always tweaking their models, and sometimes they go through temporary bouts of stupidity to the point where they’re almost completely untrustworthy.

This post just exposes the reality vs the propanganda that the venture capitalists of the AI companies try to feed us. Thanks for restoring my faith in humanity.

If you are a developer You can use AI but you MUST take some time writing code by yourself so You can learn or at least do not loose your habilities, specially if you are just starting. If don’t you will not be able to check the generated code.

Simply put, AI is really great for a quick bad answer. But it will always give you dozens of silent assumptions that you have to parse through and figure out if they are valid or not. It’s generally terrible with context.

AI is just the way to go …Mr and Mrs Stackoverflow please your time is far spent .you will soon go oblivion .who even uses you guys now