Study finds async AI coding agents drive far bigger output gains than autocomplete
NathanpmYoung · x · 2026-07-25
A figure from a study on AI coding tools compares productivity gains across three generations of tooling:
- Autocomplete (e.g. GitHub Copilot at launch)
- Sync agents (e.g. Claude Code run locally)
- Async agents (e.g. OpenAI Codex run remotely)
The chart shows the biggest gains from asynchronous agents, with roughly:
- 17.3× more lines of code
- 3.9× more files touched
- 2.8× more commits
- 2.5× more pull requests
- 1.5× more repos touched
- 1.3× more releases
The takeaway is that more agentic workflows appear to correlate with much larger throughput changes than classic autocomplete.
More from coding & agent
- OpenAI ships GPT-6.1 Terra; Brockman says its computer use now beats his own speed — gdb · 2026-10-27
- Kolab Benchmark Uses Pro Screen Recordings to Show AI Agents Overestimate Collaboration Quality — minnesotanlp · 2026-10-27
- Dev argues Claude burns far more tokens than GPT models for the same coding work — haider1 · 2026-10-27
- Stop building agents that can't change models: stable contracts, external state, real evals — rseroter · 2026-10-27
- Cheap models can cost more in agent harnesses than frontier ones, researcher warns — vasuman · 2026-10-27
- Kanzo: A 5k-Line Rust Kanban Binary With MCP for Claude Code and Codex — allmudi · 2026-10-27