ChatGPT Work's agent now runs in picture-in-picture

OpenAI shipped ChatGPT Work, a new agent living on web, mobile, and desktop, and gave its Computer Use feature a visible upgrade: a picture-in-picture mode so you can actually watch the agent operate a browser — clicking buttons, filling forms, scrolling — instead of staring at a spinner. Execution is faster too.

Under the hood it's powered by the new GPT-5.6 ("Sol") models, which OpenAI calls its "most polished collaborator yet," with stronger computer use and design judgment that let it inspect, refine, and deliver ready-to-use results rather than just draft them. A new "ultra" mode coordinates multiple agents across parallel workstreams for heavier tasks.

The desktop app also folds in Codex — edit Markdown and code inline, and review GitHub pull requests without leaving the window.

Why it matters: watching an agent work isn't a gimmick — visibility is what makes people trust it enough to hand over real tasks.