METR study: devs were 19% slower with AI but believed they were 20% faster
AlexTensor · x · 2026-10-31
Citing METR's early-2025 randomized study, the author notes that 16 experienced open-source developers working in their own repositories took 19% longer to finish tasks when using AI, yet afterward estimated it had made them 20% faster. The takeaway: even for experienced developers, and even in the domain where AI should deliver the most gains, self-assessing AI's productivity benefit is highly unreliable — subjective perception can be the exact opposite of measured reality.
Related event: METR Study: AI Tools Made Devs 19% Slower While They Felt 20% Faster(2 posts)→
More from coding & agent
- RAG evals can fail even when your system is correct — leave-one-out fix — Al_Grigor · 2026-10-31
- "Coding is becoming gardening": guardrails are stakes, review is pruning — sebpaquet · 2026-10-31
- VS Code Automations lack git commit triggers, leaving GitLab users stuck — Ill-Commission-2738 · 2026-10-31
- Free 7-week production RAG course hits 10k stars on GitHub — mdancho84 · 2026-10-31
- Tokenmaxxing may be ending: Anthropic's S-1 shows ~80% of revenue billed per token — HarperSCarroll · 2026-10-31
- 7 free Claude Skills that fix the cookie-cutter look of AI-generated pages — socialwithaayan · 2026-10-31