Founder says Codex keeps missing explicit Markdown context in an AI ops command center
No-Bus2109 · reddit · 2026-07-25
The author, who runs an AI software company, describes building a command center that aggregates Slack, Notion, email, Google Ads, and Meta Ads, with a decision queue that escalates approvals and lets the agent learn from prior decisions.
He says the system worked well with Claude Fable, where iterations usually fixed issues. After switching to Codex because of credit limits, the experience deteriorated: Codex repeatedly misses information that is explicitly present at the top of the Markdown files, apologizes, and often fixes one thing while breaking another despite the documentation and traceability being in place.
More from coding & agent
- A step-by-step recipe from supervised learning to agentic world modeling — cwolferesearch · 2026-07-25
- HarnessRouter launches with a single API for shipping AI agents in apps — ycombinator · 2026-07-25
- How I built an AI research engine with Claude Opus 5 and web sources — eptwts · 2026-07-25
- OpenHands adds Kimi Code support, metrics, and self-hosted fixes in a community release — rajistics · 2026-07-25
- OpenHands SDK 1.37.1 lets MCP servers add tools while a session is running — rajistics · 2026-07-25
- OpenHands SDK 1.37 adds persistent memory across sessions with MEMORY.md — rajistics · 2026-07-25