Understanding ChatGPT Work
- ID
- 19694
- Status
- summarized
- Published
- 31 Aug 2026, 7:59 AM
- Fetched
- 01 Sep 2026, 9:09 AM
- Provider
- Simon Willison
- Category
- developer-ai
- Original URL
- https://simonwillison.net/2026/Aug/30/understanding-chatgpt-work/
- Source URL
- https://simonwillison.net/atom/everything/
Summary
- Score
- 7.5
- Created
- 01 Sep 2026, 9:09 AM
- Tags
- Audience
- developersvibe_codersai_agent_usersai_ml_learnerssaas_founders
What happened
Simon Willison dissects ChatGPT Work, which OpenAI launched July 9th and has rapidly iterated since. It comes in two flavors: Work Cloud (via chatgpt.com/mobile) and Work Local (the desktop app formerly called Codex). Work is gated to $20/month+ subscribers and adds features absent from regular Chat: model selection between GPT-5.6 Sol, Luna, and Terra with six reasoning levels, a code execution environment with internet access, a headless Chrome browser, a persistent cross-session filesystem, ChatGPT Sites publishing, sub-agent sessions, and scheduled prompt automations.
Why it matters
If you're paying $20/month for ChatGPT, you should know Work Cloud gives you a persistent filesystem, headless Chrome, internet-connected code execution, and sub-agent orchestration that Chat does not—features that overlap with what many people currently build with separate agent frameworks. Decide whether to route agentic tasks through Work Cloud instead of stitching together your own tooling, and note that model selection differs between Chat and Work, which affects which reasoning levels you actually get access to.
Discussion angle
Compare what Work Cloud's persistent filesystem, headless Chrome, and sub-agent sessions cover versus what you still need a separate agent framework (LangChain, CrewAI, etc.) for—does this shrink the build-your-own-agent stack for $20/month subscribers?