Unpacking ChatGPT Work: the Agent for a Billion Users

OpenAI’s ChatGPT Work agent, released in July 2026, consolidates years of experimental products into a unified agent for knowledge work. It runs on the Codex harness inside a persistent cloud computer, integrates with hundreds of external services via plugins, and can proactively suggest tasks based on a user’s calendar, email, and memory. The product is a preview of how ChatGPT’s billion weekly users will eventually use the app, as Chat and Work are set to merge by the end of the year. The article reconstructs how Work’s memory, proactivity, scheduling, browser use, plugins, and tooling function under the hood, drawing on direct experimentation and conversations with the agent. It highlights architectural choices (e.g., a cloudy computer with structured continuity rather than a fully sovereign agent) and identifies unresolved tensions around cloud-local syncing, agent autonomy, and plugin discovery. These details matter because they reveal the design philosophy behind the mainstream AI agent experience and the challenges OpenAI must solve before Work becomes the default interface for all ChatGPT users. Key technical points from the report are summarized below. <br><br> <b>Persist<