OpenAI has shipped workspace agents inside ChatGPT, built on Codex, designed for team environments. The demo shows a single agent handling the full lifecycle of a software request: policy check, approval routing, and IT ticket creation, end to end, without human handholding at each step.
The architecture matters more than the output. Codex is doing the reasoning and action-chaining across what would normally be three or four separate tools and human touchpoints. That compression of workflow steps into one agent loop is the technical argument OpenAI is making here.
Watch the full walkthrough to see exactly where the agent makes decisions versus where it escalates, and how it structures the IT ticket output. Those design choices reveal the boundaries of what workspace agents can and cannot own autonomously right now.
[WATCH ON YOUTUBE →]