Researcher sweeps months of ideas with agents, using Codex+Claude Code error-check loops
burny_tech · x · 2026-10-25
A researcher reports that concrete, hillclimbable research ideas can now be mostly executed by agents. His workflow: sweep months of noted ideas, then have Codex and Claude Code double-check each other in a loop — verifying specs were followed, plans fully executed, scientific rigor, hallucinations, correct library/math/empirical grounding, security, dependency versions, and overall technical coherence. It catches errors, but he still inspects critical projects closely himself.
More from coding & agent
- Agents still bad at dependent types: their instincts skew imperative even when writing Lean — hargup13 · 2026-10-25
- LangChain's Harrison Chase teases enterprise answer to OpenClaw, offers early access — hwchase17 · 2026-10-25
- Catalogue of 48 programming languages designed for AI agents maps three competing camps — QuixiAI · 2026-10-25
- MCP Sage pipes whole codebases into Gemini 2.5 Pro for second-opinion code reviews — modelcontextprotocol · 2026-10-25
- LangChain founder: AI content studio now makes all our diagrams, videos and PPTs — hwchase17 · 2026-10-25
- Microsoft open-sources Skill Recorder: record a task once, your Copilot agent learns to do it — udmrzn · 2026-10-25