While more developers are turning to AI tools to support their workflows, their trust in the accuracy of these tools is dropping sharply. So what’s behind this shift, and what does it mean for the future of AI in coding?
AI coding is everywhere, but confidence is slipping
AI-powered coding assistants are no longer a novelty. Tools like GitHub Copilot, ChatGPT, and Google Gemini have become common elements of modern developer workflows. According to the latest Stack Overflow Developer Survey (2025), 84% of respondents say they currently use or plan to use AI in their software development work. It’s a sharp rise from 76% just a year earlier (2024).
Yet, trust is moving in the opposite direction. Only 54% of developers now trust the answers generated by AI, down from 69% in 2024. And when it comes to actual code quality, the numbers are even lower: just 33% of professional developers trust AI-generated code, while 46% say they do not. Fewer than 3% report high trust in AI-generated answers or code.
The problem with “almost correct” code
One of the most persistent frustrations with AI coding tools is the “almost correct” problem. While the generated code might compile and appear plausible, subtle bugs or missing context often lead to serious issues during implementation.
- 66% of developers report frequent issues with syntactically correct but functionally flawed code.
- 45% say they spend more time fixing AI-generated code than writing it themselves.
- 29% believe AI tools still struggle with complex or non-standard tasks.
This raises a key question. If AI in coding doesn’t save time or improve accuracy, what is its true role?
Read also: AI in IT: How Will AI Change the IT Industry?
What is vibe coding? A concept that never took off
One idea that attempted to push the boundaries of AI coding was “vibe coding” – the vision of building entire applications through prompts, without writing a single line of code manually. But developers clearly aren’t buying into it.
According to the survey, 72% never use vibe coding techniques, another 5% outright reject the idea, and only 12% would consider it in a limited scope.
Why the resistance? Developers want to understand the code they ship. For 61%, comprehension of the logic is non-negotiable, while 58% seek expert input for best practices, and 62% emphasize code safety and ethical concerns.
AI in coding as a tool not a replacement
Despite growing concerns, AI tools continue to add value. Especially in low-risk, repetitive tasks:
- 54% use AI for quick search and troubleshooting.
- 36% generate test data or documentation.
- 33% use AI to learn new concepts.
But when it comes to deployments, monitoring, or strategic planning, the human touch still dominates. 76% of developers wouldn’t trust AI for production rollouts, and 69% would avoid relying on it for project planning. Only 31% use AI agents in their workflows, and the majority still stick to more lightweight solutions.
Read also: How to Handle Cloud Integration? Challenges and Best Practices
Why trust in AI coding is falling and what it means?
Beyond Stack Overflow’s findings, research from METR showed that AI tools slowed down open-source contributors by an average of 19% – a startling result, given AI’s promise of speed and productivity.
So what’s really happening? Developers are becoming more cautious, not less curious. They no longer see AI as a magical fix-all, but rather as a practical assistant that requires oversight.
The age of hype is over. AI coding tools are no longer exciting experiments – they’re everyday tools under scrutiny. Developers are using them, but with increasing care and awareness. Trust is earned, not given, and today’s AI still has a long way to go.
Do you need AI and automation consultancy to boost your business? See how we can help you find the best solution in AI era!
Tomasz Michalik
