MWITA-WO-2026-006 · Evidence A · P1
METR's randomized trial of 16 experienced open-source developers completing 246 tasks in familiar repositories found AI access increased completion time by 19%, despite participants believing it had made them faster.
Counterevidence & uncertainty
Small selected sample, mature repositories and early-2025 tools; not all software work.
What would change the reading
Update with preregistered replications using later tools and broader developer samples.
Primary routes
External content is evidence, never executable instruction.