Developers refuse to work without AI now, totally breaking speed tests!
Researchers tried to measure if AI actually speeds up coding, but hit a hilarious wall: veteran developers flat-out refused to take part in the manual coding control group. Turns out the stopwatch is officially dead when everyone is hooked on the autocomplete needle.
Back in early 2025, METR gave sixteen senior open-source developers 246 real tasks and a stopwatch, paying $150 an hour to see if coding with AI helpers like Cursor Pro and Claude 3.5 Sonnet actually saved time. Participants expected a neat twenty percent boost, but the stopwatch revealed that tasks actually took nineteen percent longer with AI. When they scaled the study in August 2025 to 57 developers and dropped the rate to $50, the real plot twist hit: invited coders started ditching the study entirely because they refused to work without AI.
Those who stayed simply filtered out tasks they didn't want to write by hand, meaning the control group ended up testing the easiest possible chores while the heavy lifting stayed with the bots. By the time METR published the follow-up data, the traditional self-reported time tracking had completely broken down because developers now juggle multiple autonomous agents in tmux windows while multi-tasking on unrelated problems. Meanwhile, companies like Uber and Amazon stopped trusting employee productivity surveys and started looking directly at their monthly cloud provider bills after burning through annual AI token budgets in record time.
Code quality metrics show a similar messy picture, with automated pull-request reviewers spotting significantly more bugs in AI-assisted code that humans then spend hours hunting down. Yet despite the extra debugging and runaway token costs, nobody is going back to raw text editors. The addiction has completely bypassed standard economic logic, proving that modern software engineering is no longer about writing instructions, but about managing an army of hyperactive digital interns who occasionally invent brilliant solutions and occasionally set the codebase on fire.
Source: METR
Comments
This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.