Developers may be too dependent on AI to measure coding productivity cleanly

The AI-coding productivity debate just hit a measurement problem. TechCrunch reports that METR wanted to repeat earlier research comparing open-source developers working by hand versus with AI, but could not because many developers refused to work without AI even for a limited study. That matters because perceived productivity and measured productivity are diverging. METR's earlier work found developers felt faster but were slowed by steering, waiting, and fixing errors; a newer survey found workers self-reporting much larger gains. The article also ties in rising token spend, Amazon's reported Kirorank shutdown, Uber's AI-budget concerns, and claims from code-review vendors that AI-generated code creates more defects. The useful signal is not anti-AI; it is that software teams need maintenance, review, and cost metrics rather than treating agent usage as proof of productivity.

Key details: TechCrunch, METR, May 29, 2026, AI coding productivity, developers refused no-AI tasks, Amazon Kirorank, Uber AI budget, 1.7x more problems claim.

Continue swiping for more AI Brief stories.