Best coding agent scores just 38.8% on a new test, failing 6 of 10 tries
JFPuget · x · 2026-09-15
Jean-François Puget (NVIDIA) amplified a post noting the best coding agent scores only 38.8% on a new test — meaning it fails about 6 out of 10 tasks. The point: real-world agent success rates remain far below the hype.
More from coding & agent
- Chatbot injects 252k-token prompt every turn instead of RAG, dev asks to be talked out of it — louay_Sallakho · 2026-09-15
- Agent browser stacks break at logged-in flows: session restore is the real problem — Jocelyn-Skyquill · 2026-09-15
- In the Vibe Coding era, React is becoming a programming language — yihui_indie · 2026-09-15
- MCP vs A2A: two layers of one stack, and why orchestrators still do all the real work — Diarnstinc_Crew_6510 · 2026-09-15
- Open-source ProxyPin MCP Workbench lets AI assistants inspect mobile app traffic — Designer-Subject8560 · 2026-09-15
- Dev builds MCP server so agents analyze his X account data before drafting posts — DarkyZ_ · 2026-09-15