Codex Founder Tibo: Partial Service Interruption for Codex
Tibo, the founder of Codex, disclosed that OpenAI's Codex is currently experiencing partial service outages. The team is investigating the issue but has not yet announced the specific cause or recovery time.
The problem arose shortly after Codex launched new features such as Memories and Chronicle, with a significant increase in system load and backend agent call frequency. There is external concern about whether this is related to resource scheduling or the stability of the new features.
It is not uncommon for similar developer tools to experience brief interruptions during rapid feature iteration phases. Previously, several AI coding tools have experienced stability fluctuations after introducing long context and automated agents.
Source: Public Information
ABAB AI Insight
This type of "partial outage" in the current evolution stage of AI tools is essentially an external manifestation of architectural pressure rather than a single point of failure. Codex is transitioning from a model that "passively responds to requests" to a "continuously running multi-agent system," where requests are no longer discrete but persistently generated in the background, directly altering the load structure of computational resources.
The key issue with features like Chronicle lies in "implicit calls": users do not explicitly request, but the system continuously takes screenshots, performs OCR, summarizes, and writes to memory. This mode transforms controllable token consumption into invisible continuous consumption, easily triggering rate limits, queue congestion, or even service degradation.
From a broader perspective, this reflects that AI products are moving from "tools" to "infrastructure." Once in the infrastructure phase, the importance of stability will surpass the model's capabilities. The deeper the developers' reliance on AI, the smaller the margin for error; any interruption will directly impact production processes, which is why "availability" has become a core competitive metric in the post-cloud computing era.
On a deeper level, this fluctuation reveals the structural contradictions faced by current AI companies: on one hand, pushing for stronger automation and contextual awareness, while on the other hand, still constrained by computing costs and scheduling capabilities. The industry has yet to find a stable balance between "continuous agentization" and "controllable costs."