OpenAI launches Workspace Agents in ChatGPT for team collaboration

What changed

OpenAI rolled out Workspace Agents inside ChatGPT on April 23, 2026. Organisations build a custom agent once, share it across the company, and have it run autonomously across ChatGPT plus Slack, Gmail, and Salesforce. Behavioural commitments in the launch blog: gather context from connected systems, follow codified team processes, ask for approval on high-stakes actions, and keep work moving across tools. Free in research preview until May 6, 2026, then credit-based pricing on Business, Enterprise, Edu, and Teachers plans. Plus and Free users do not get access at launch.

Why it matters

Workspace Agents runs on OpenAI's Codex platform and surfaces AgentKit, introduced in October 2025, as a product rather than raw SDK. AgentKit's Agent Builder, Connector Registry, and ChatKit are now bundled behind a single shareable-agent abstraction. If your stack already calls AgentKit directly, the team-sharing, approval-gate, and cross-tool orchestration logic is now off-the-shelf; integration code you wrote three months ago may be redundant. OpenAI is also explicitly framing this as the successor to custom GPTs, with one-click GPT-to-Workspace-Agent conversion promised soon.

What to watch for

Watch the credit-based pricing model that kicks in after May 6: per-agent-run economics will determine whether this scales for high-frequency workflows or stays a low-volume tool. Watch the connector list past Slack, Gmail, and Salesforce; AgentKit's Connector Registry depth is the real ceiling on what Workspace Agents can do. And watch whether AgentKit-direct usage retains a pricing or capability edge over the packaged surface, because that decides whether custom builds remain worth the code.