Agentic AI growth is colliding with the economics of computing.
San Francisco
OpenAI is restoring a five-hour usage limit for ChatGPT Work and Codex on ChatGPT Plus accounts, reintroducing a restriction that had been temporarily removed while users operated primarily under weekly quotas. The change takes effect on August 25 and applies to two of the company’s most compute-intensive experiences: Work, designed for long and complex professional projects, and Codex, its specialized software engineering agent. Users who reach either the five-hour threshold or their weekly allowance will have to wait for capacity to reset or use additional credits. The measure highlights a growing challenge for the artificial intelligence industry as increasingly autonomous agents require far more computational resources than conventional chatbot interactions.
The restriction does not mean that ChatGPT itself can only be used for five hours. It specifically affects the usage structure surrounding Work and Codex for Plus subscribers, while ordinary Chat interactions remain governed separately. Work is designed to perform multistep tasks across files, applications and workflows, producing finished documents, spreadsheets, presentations and other deliverables. Codex performs similarly intensive operations in software engineering, where an agent can inspect code, modify files, execute tests and repeatedly evaluate its own progress.
OpenAI has linked the restored limit to managing computational demand while preserving relatively generous weekly usage. The distinction is important because agentic systems can remain active across lengthy workflows, generating substantially greater resource consumption than a conventional question-and-answer exchange. A single complex assignment may require an agent to plan, execute, verify and revise multiple operations before producing the final result. As adoption grows, controlling when and how that workload reaches data centers becomes an increasingly important part of operating these services at scale.
The decision also exposes one of the central economic questions surrounding the next generation of artificial intelligence. Technology companies are encouraging users to move from asking AI for answers to delegating complete projects, yet every additional layer of autonomy can increase computational demand. Usage limits therefore reveal something larger than a subscription policy: the emerging agent economy must balance capability, accessibility and infrastructure costs. The future of AI may depend not only on how intelligent agents become, but on how efficiently millions of people can use them simultaneously.
Beyond the news, the pattern. / Más allá de la noticia, el patrón.