Codex Runs 21 Hours, Ships Feature

The real signal here is not the duration, it’s the system design. If an agent can keep going through repeated rejections, re-evaluate, and eventually land clean code, that’s a stronger story than raw speed.

• –This is anecdotal, not independently verified, but it suggests Codex can sustain unusually long unattended sessions when the task loop is engineered well.
• –The O.M.A.R. gate is the interesting part: deterministic checks turn a fragile agent into a self-correcting pipeline.
• –The post reinforces a broader pattern in agentic coding: durability comes from guardrails, not from hoping the model “just knows” the right answer.
• –For teams, the practical lesson is that autonomous coding is a systems problem as much as a model problem.
• –The Claude Code comparison is mostly rhetorical, but it highlights the current competitive axis: not just code quality, but how long an agent can operate safely without intervention.