OpenAI Launches Workspace Agents to Automate Team Workflows

OpenAI introduced workspace agents in ChatGPT for teams, a cloud-hosted agent capability that runs long-running, multi-step workflows on behalf of organizations. The feature is powered by `Codex` and available as a research preview to Business, Enterprise, Edu, and Teachers plans. Workspace agents are designed to gather context from company systems, follow processes, request approvals, and act across tools such as Slack and Gmail. OpenAI says workspace agents are an evolution of GPTs and will remain convertible from existing custom GPTs.

The agents run in the cloud and are intended to operate autonomously across handoffs and internal systems while respecting organizational permissions. OpenAI pairs the rollout with updates to its Agents SDK that include sandboxing and an in-distribution harness for frontier models. Key technical capabilities called out by OpenAI and reporting outlets include:

• •multi-step workflow execution and long-horizon task handling
• •context aggregation from files, notes, and integrations across tools
• •configurable approval flows and permission boundaries for actions
• •convertibility from existing GPTs into workspace agents

• •Workspace agents are guided-creation experiences inside ChatGPT, letting teams describe recurring workflows and turn them into agents quickly.
• •The Agents SDK now supports sandbox providers so agents can execute code or interact with files in controlled environments, limiting blast radius.
• •The in-distribution harness provides a controlled runtime for testing agents on frontier models before full production deployment.

Workspace agents represent OpenAI moving from single-session assistants to genuinely agentic, team-shared automation. This follows industry momentum around agentic AI, where startups like OpenClaw and competitors such as Anthropic have emphasized agents that do things rather than just assist. For enterprises, the combination of Codex-powered automation plus SDK sandboxing addresses two practical needs: the ability to automate cross-tool workflows and the need to reduce unpredictable or unsafe behavior when agents take actions.

This product also signals a consolidation in OpenAI's product surface: custom GPTs are being positioned as legacy primitives that will be upgraded or converted into workspace agents. That matters operationally because teams that invested in GPTs will need migration paths and governance controls. The research preview model and the announced free window until May 6, 2026 give teams time to pilot before credit-based billing begins.

Adoption friction will come from integration and governance. Successful deployments will require clear role-based permissions, robust sandboxing of connectors, and reliable observability for audit and debugging. Monitor the Agents SDK release notes and admin controls in ChatGPT Business for the specifics of permission models, connector whitelists, and conversion tooling from GPTs to workspace agents.

Workspace agents are a practical step toward operational agentic AI for teams. Practitioners should evaluate them for automating recurring cross-tool workflows, but plan for governance, testing in sandboxes, and migration of existing custom assistants.