Researchers say developers are now reluctant to work without AI coding tools, even as code quality and maintenance risks remain unresolved.
AI coding tools have become part of daily software work. Now, some researchers are asking whether the industry has moved faster than the evidence.
TechCrunch reported that developers are increasingly unwilling to take part in studies where they must work without AI. That matters because it makes the real effect of these tools harder to measure.
The issue came up in work by METR, an AI research group. METR said in February 2026 that a new developer productivity study had become unreliable because many programmers did not want to work without AI coding tools.
That is a shift from earlier research. METR’s 2025 study found that AI tools caused experienced open-source developers to take 19% longer on selected tasks. The researchers later warned that newer results may be affected by selection bias, because the most active AI users may avoid studies that limit the tools.
There is a positive case. Developers say these tools help with drafts, tests, documentation and routine fixes. In a May 2026 METR survey of 349 technical workers, respondents said AI coding tools raised the value of their work by a median of 1.4x to 2x.
But self-reporting is not the same as proof. METR also said survey answers can overstate real productivity gains. The bigger concern is not whether code appears faster. It is whether teams can maintain it.
A separate April 2026 paper on AI-generated code in real projects found that more than 15% of commits from every AI coding tools studied introduced at least one issue. The paper also found that 22.7% of tracked AI-introduced issues were still present in the latest version of the repository.
That does not mean developers will stop using AI. They probably will not. It means companies may need stronger reviews, clearer ownership and better testing before they treat generated code as finished work.
The AI Decode has also covered how AI infrastructure is changing job demand. The same wider shift is now hitting software teams. Speed is useful. But the real test is whether the code still makes sense six months later.
