METR Study: Developers Felt 20% Faster With AI, Actually Took 19% Longer
alifcoder · x · 2026-11-20
A 2025 METR study had 16 experienced developers use AI on familiar coding tasks. Measured results: they took 19% longer — yet afterward they estimated AI had made them 20% faster.
This is the productivity trap: a quick first draft feels like progress, but review, corrections, and integration count too. The gap between perceived and actual speed is a warning for anyone relying on AI coding assistants.
More from coding & agent
- A copy-paste 'reporting contract' for AGENTS.md to fix GPT-6.2's poor interface — daniel_mac8 · 2026-11-20
- Florian Roth: open source vulnerability surge reflects LLM-driven scrutiny, not worse code — cyb3rops · 2026-11-20
- lattedit turns your browser into a LaTeX editor that shows Claude Code, Codex and Cursor editing live — cneuralnetwork · 2026-11-20
- Fable 5.2 filmmaking playbook plus 12 repos that give coding agents design taste — GCWebDesigner · 2026-11-20
- Bolna's Symphony event breaks down India's voice-agent stack: telephony, voice models, ASR — prasanna_says · 2026-11-20
- Which AI agent sandbox is best? Daytona vs e2b for dev and scale — Ill_Acanthaceae_2618 · 2026-11-20