Developers refused to work without AI: a study revealed a dependency paradox

In February 2026, the METR research laboratory encountered an unexpected problem. It attempted to replicate a 2025 study that measured how AI affects programmer productivity. But the experiment failed — developers simply refused to participate without their usual AI tools, even for a controlled laboratory test.

What Happened in the Lab

The original 2025 study showed an important result: developers work 30–50% faster with AI assistants like GitHub Copilot and ChatGPT. This was a major contribution to understanding AI's impact on productivity and business processes in IT. A year later, METR decided to repeat the same experiment to understand whether anything had changed over the past year. But when the laboratory invited programmers to participate in the research, something unexpected happened. Developers refused to write code without their usual AI tools — even for a short test in controlled conditions. This was not a polite refusal; people simply did not want to turn off Copilot, ChatGPT, and other assistants.

Why People Refused

Over the course of one year, AI became not just a convenient tool, but a necessity in everyday work. Developers had completely adapted to constant autocomplete, smart suggestions, and debugging assistants. They had forgotten how to work the old way — without this digital background. Moreover, study participants explained their refusal simply: without AI, their productivity seemed to have dropped, and work felt incomplete. This is a classic case of technological dependency: when a tool becomes so integrated into a process that work seems impossible without it.

• Adapted to constant autocomplete and code suggestions
• Without AI, work feels slow and inefficient
• Forgot how to solve tasks independently
• Fear losing productivity, even for a short period

The Dependency Paradox

The inability to repeat the study itself is a significant result. It shows that over one year, AI integration into development became so deep that developers simply refuse to work without it, even for scientific purposes. But this raises a dangerous question for us: if people are completely dependent on AI and cannot (or will not) work without it, does this mean AI actually makes them better? Or is it just habit? And most importantly — what will happen to developers' skills if they rely on AI for 80–90% of their work?

"We could not conduct the study because people simply refused to

participate without their AI tools," said a METR representative.

What This Means

The paradox is that AI tools, created to boost productivity, have created complete dependency. Developers cannot (or will not) work without them, but this does not necessarily mean they are better developers. It may mean that they have become dependent on AI rather than becoming smarter at solving problems independently.