Paper: general coding agents beat purpose-built data agents by up to 37 points
RishiBommasani · x · 2026-09-09
A new paper co-authored by Ion Stoica reviews two years of data agent evolution: general coding agents now significantly outperform hand-designed data agents by up to 37 points with 4× fewer turns. The authors break down which systems research problems remain for building the next frontier of data agents as better models eat the stack.
Related event: Study: general coding agents beat specialized data agents by 37 points(4 posts)→
More from coding & agent
- Yoav Goldberg proposes a new agent eval: locating real GitHub artifacts via API — yoavgo · 2026-09-09
- OpenAI's Navier-Stokes proof reportedly used ~10,000 coordinating AI agents — mark_k · 2026-09-09
- Open-source Aurora: Go LLM gateway fork claims 55x LiteLLM speed and fixes OpenCode session lockout — entitybtw · 2026-09-09
- Can agents resolve vague human references like 'the change that broke token refresh'? — yoavgo · 2026-09-09
- Dev open-sources local prompt optimizer on DSPy+Ollama, boosting llama3.2 3B from 39% to 89% with GEPA — MortisAndTen · 2026-09-09
- Dev picks Amp over Codex for thread-focused agent UI — carlosdponx · 2026-09-09