Unpacking ChatGPT Work: the Agent for a Billion Users
Source
Shlok Khemani
Author
Shlok Khemani
Date
Key takeaways · AI-distilled
Work runs on the Codex agent harnessThe scaffolding around a model that turns it into a working agent — the loop, the tools it can call, and the rules for when to stop.Full definition → inside a persistent cloud microVM (8 CPUs / 20GB RAM on Pro): the workspace syncs to persistent storage and is restored onto fresh isolated VMs, so the machine changes but working state carries over.
Each task gets a /workspace/scratch directory with real computer freedom — install dependencies, keep databases, write scripts — though the AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → has far less sovereignty over its machine than an OpenClaw-style always-on box.
Web and mobile always run in the cloud; the desktop app adds a local mode that is essentially Codex with the coding UI stripped away, and local tasks do not sync to web or mobile.
Chat and Work sit side by side today but will merge by end of year (confirmed by Brockman) — Work is a preview of how ChatGPT's billion weekly users will use the app.
Artifacts are first-class output: sheets, docs and slides in interactive viewers, plus hosted Sites — web apps the agent builds, shares by URL, and keeps updated.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
MCP — The Model Context Protocol — an open standard that lets any AI assistant plug into any tool or data source without custom integration code.
agent harness — The scaffolding around a model that turns it into a working agent — the loop, the tools it can call, and the rules for when to stop.
Why it matters
It explains concretely how ChatGPT Work implements persistent memory (Personal context windowThe maximum amount of text a model can consider at once — its working memory for the current conversation or task.Full definition →/Library), heartbeat-style scheduled automations, a separately hosted browser service with persistent login, and an MCPThe Model Context Protocol — an open standard that lets any AI assistant plug into any tool or data source without custom integration code.Full definition →-based Plugin Directory — details useful for anyone designing similar agent architectures.
Key quotes
“On July 9th, OpenAI released ChatGPT Work, their agent product for knowledge work.”
“None of this should detract from the fact that Work is an impressive, ambitious, yet underrated launch.”
