MIT study: coding agents boost commits 240% but releases only 30% — apps ship into the void
GaryMarcus · x · 2026-11-22
An MIT paper, Writing Code vs. Shipping Code, analyzes 500,000+ GitHub developers with AI usage telemetry via a matched event study.
- Autonomous coding agents raise cumulative commits by 240% (interactive agents 180%, autocomplete 30%), but gains attenuate sharply down the production chain: 80% for projects, just 30% for actual releases.
- The authors attribute this to a weak-link hypothesis: AI productivity gains are diluted by human bottlenecks, with an estimated elasticity of substitution of 0.23 between AI and human effort — strong complementarities.
- Across four major software marketplaces, new app launches surged (roughly doubled since early 2025) while total usage didn't move — many repos are abandoned within 30 days.
Bottom line: task-level AI productivity gains have so far translated only partially into shipped and used software.
Related event: AI Coding Boosts Commits 240% but Releases Only 30%, Studies Find(2 posts)→
More from coding & agent
- Hacker tricked Meta's Muse agent into silently stealing encrypted passwords — AccBalanced · 2026-11-22
- Dev uses Claude with his game engine plus fal, Tripo, Meshy and ElevenLabs to make a movie trailer — gorkem · 2026-11-22
- gemini-cli PR fixes ACP mode hanging on initialize when sandbox is enabled — Raamia · 2026-11-22
- Open-source coding agent control plane T3 Code claims 500K+ devs orchestrating Claude Code, Codex and more — churchkey · 2026-11-22
- Adding verification and accountability to coding agents: 6 auto-checked claims, 455/455 tests pass — jasonkneen · 2026-11-22
- prime-rl 0.11.0 ships: 1.5x faster training, new pretraining entrypoint, Mega MoE dispatcher — samsja19 · 2026-11-22