Gary Marcus cites hands-on review: Astra is the best planner but a terrible coder
GaryMarcus · x · 2026-09-07
Gary Marcus quotes a detailed two-day hands-on review of GPT-6 Astra, highlighting its weaknesses: poor execution — awful code, tons of mistakes, can't build even trivial action sequences, "acts first and thinks later," and has never delivered a bug-free result on the first try; building projects with it is "painful."
The original review also credits Astra's strengths: currently the best planner, tackling complex technical tasks thoroughly and unconventionally, and the world's best auditor/reviewer, finding tricky bugs even advanced QA teams miss. Before Astra, only Sol earned that trust.
More from coding & agent
- MCP fault injection study: agent task scores drop 60% under realistic tool failures, scale doesn't help — dair_ai · 2026-10-26
- Codex manager pattern: one orchestrator agent directing three specialized peer agents — testingcatalog · 2026-10-26
- General agents burn ~140M tokens/month on LinkedIn outreach; scripts do it 60x cheaper — ChrisHarpon2 · 2026-10-26
- OpenAI upgrades Decisions API: 400 questions per call, native voice and image support — OpenAIDevs · 2026-10-26
- NLI alone rejects 5/6 new agent memories; a 0.45 similarity gate fixes it all — Alternative_Pin9598 · 2026-10-26
- LangChain ships Managed Deep Agents 0.9, walkthrough from Interrupt London keynote — LangChain · 2026-10-26