Codex Head Tibo Reveals Iteration Cycle: Ship-Feedback-Reset Repeatedly
OpenAI Codex head Tibo described the team's current work rhythm: Ship → Feedback → Ship → Feedback → Ship → Identify Issues → Reset → Yay → Ship… a continuous cycle of product improvement.
He specifically mentioned a recent caching issue, which has been resolved by clearing all caches and redeploying. This cycle is the norm for Codex's ongoing optimization.
Source: Public Information
ABAB AI Insight
Tibo, as the core head of Codex, has previously been deeply involved in the iteration of OpenAI's code generation models from the early Codex to the built-in coding capabilities of ChatGPT. This public sharing of the "Ship-Reset" cycle continues OpenAI's internal culture of rapid experimentation and high-frequency deployment, similar to the early GPT series which adopted a "fail fast - clear cache and restart" engineering path.
On the capital front, OpenAI is investing significant engineering resources into parallel optimization of Codex for mobile and desktop, rapidly enhancing the model's performance in real developer scenarios through frequent cache resets and feedback loops. The motivation is to seize the mobile coding and agent development window while accumulating vast amounts of real usage data to form a training flywheel.
Similar to the weekly multi-version iterations of Cursor and GitHub Copilot, as well as the rapid fix cycles of Anthropic Claude's coding features, OpenAI Codex is currently in a high-frequency iteration phase transitioning from a basic completion tool to a full-scene intelligent coding platform.
Essentially, this represents a technological replacement: traditional code development relies on stable but slow deployment processes, while the Codex team shifts engineering practices from "pursuing a perfect one-time release" to "continuous real-time evolution" through high-frequency cycles of "Ship-Feedback-Reset," allowing AI products themselves to become dynamic learning systems, accelerating AI iterating AI, and pushing developer tools from static software to living intelligent agents.
ABAB News · Cognitive Law
True iteration is never afraid of Reset.
The cost of clearing the cache brings the speed of the next leap.
Products are not built perfectly once, but become stronger through continuous redoing.