Why it’s still open
Experienced developers took longer with early-2025 AI tools yet believed they were faster; people predict AI help will speed them up even when it does not, and overestimate their scores with AI. The gap between felt and measured progress may be what keeps the loop going — but its mechanism is unknown, and newer tools may be faster in fact.
What would count as an answer
Experiments separating perceived effort, perceived progress and actual time, and testing whether calibration feedback closes the gap.
Papers that tackle it
- Becker et al 2025 - METR developer productivity — 19% slower, felt 20% faster.
- METR 2026 - Developer productivity update — Newer estimates, weaker evidence, and developers refusing to work without AI.
- Yu et al 2026 - The speedup illusion — Predicted speedups where none exist.
- Fernandes et al 2025 - Smarter but none the wiser — Overestimating performance.
Methods in play
Who’s closest
Placed by the editor, v1 — corrections welcome.
Related questions
Know a paper that belongs here, or think this is framed wrongly? Suggest a paper or a correction (forms on the Atlas home).