Dev shares agentic workflow: looping Codex and Claude Code to cross-check every output
burny_tech · x · 2026-10-19
A developer describes his pilot workflow for agentic systems: looping checks over instructions and specs with Codex and Claude Code until they sign off, verifying plans were fully executed, checking for "science crimes," factual errors and hallucinations, technical coherence, correct library/math usage, empirical grounding of claims, security, and dependency freshness — all with web-search grounding. Even then, he admits, it's still not perfect, a candid look at today's agent reliability ceiling.
More from coding & agent
- Swarms ships v16 'Overclock', claims 53.7k GitHub stars and 800k developers — KyeGomezB · 2026-10-19
- Arvind Narayanan: The question shifts from 'What can AI do?' to 'What should AI do?' — soumitrashukla9 · 2026-10-19
- Dev ports 86Box to WebAssembly with Claude Code: a buildable, breakable retro PC in the browser — No_Twist_678 · 2026-10-19
- Claude Code 2.1.303 Released with CLI Bug Fixes for Fewer Crashes — ClaudeCodeLog · 2026-10-19
- Solo dev ships Stillworlds, an 11-world Mac ambient app built with Claude Code, for £5.99 — ninja81702 · 2026-10-19
- Indie dev builds Utsushi, an open-source Higgsfield clone in Go/HTMX — _playlogic_ · 2026-10-19