Researchers say developers now rely on AI coding tools, with risks

In short: Many software developers now say they will not do certain work without AI coding tools, but researchers warn this habit can hide slowdowns and create more work later.

What's going on

A research lab called METR reported in February 2026 that it struggled to repeat an experiment because many developers did not want to work without AI, even for a study. METR had tried to measure how long tasks took with and without AI help.

This matters because METR’s earlier research found a mismatch between how developers felt and what the clock showed. Developers said AI made them more productive, but in the test they often took longer overall. METR said AI could generate code quickly, but people then spent extra time guiding the tool, waiting for it, and fixing mistakes.

Since they could not rerun the test, METR later published a survey in May that relied on workers reporting their own results. Many respondents said AI made them about twice as valuable at work. Other reports in 2026 have raised doubts about those self-reports, including stories of companies tracking AI usage by counting “tokens” (small chunks of text used when you talk to AI, like charging by the word) and then seeing costs rise without clear gains.

Separate research has also warned that AI-written code can increase maintenance work later. Maintenance is the ongoing job of keeping software running and fixing bugs (like constantly patching a leaky roof).

What to watch

Expect more companies to focus on quality checks, not just speed. Researchers and some tool makers say AI output should be reviewed like work from a junior teammate, and humans still need to handle big-picture design and security.

Source: TechCrunch AI