OpenAI merges Codex into ChatGPT with 'ChatGPT Work,' an agent for finished output

ChatGPT Work represents OpenAI's clearest push yet to reposition ChatGPT from a chat assistant into an agent that ships completed deliverables. President Greg Brockman framed it as 'an agent for your most ambitious work,' emphasizing that it runs from mobile and web as well as desktop — 'no need to leave your laptop cracked open.' The feature folds the standalone Codex coding tool into the main app, unifying what had been separate surfaces.

Mechanically, the pitch is goal-to-output: a user states an objective and the agent produces spreadsheets, slide decks, or working web apps, chaining tool calls and multi-step reasoning under the GPT-5.6 family. Pro, Enterprise and Edu users get access first, with a broader global rollout following.

The move drew immediate skepticism from prominent voices. Simon Willison pushed back on the broader 'AI employees' framing the launch invites, writing that treating AI as staff is 'both disrespectful to humans and a complete misunderstanding of what these tools can do' — 'you may as well start adding Excel spreadsheets to your org chart' (1,015 likes). The competitive context is a crowded agentic-work field: Anthropic just added an in-app browser to Claude Code on desktop, and Google, Microsoft and xAI are all racing on agentic coding. François Chollet captured the vibe, noting 'it's mind-blowing how fast agentic coding has progressed in the past 6 months.' The open question is whether 'finished output' holds up on real enterprise tasks or lands as another impressive demo that stumbles on messy, real-world documents.