- Workspace agents, powered by Codex, launched in research preview on April 23, 2026 for ChatGPT Business, Enterprise, Edu, and Teachers plan subscribers.
- Agents run persistently in the cloud, execute code, retain memory across sessions, schedule recurring tasks, and integrate with Slack.
- OpenAI’s internal accounting team uses a workspace agent to complete month-end close tasks—journal entries, balance sheet reconciliations, and variance analyses—in minutes.
- Enterprise administrators can enforce per-action approval gates and access full audit logs through OpenAI’s Compliance API.
What Happened
OpenAI on April 23, 2026 introduced workspace agents in ChatGPT, a new category of persistent, cloud-running AI agents powered by Codex designed to automate complex, multi-step organizational workflows. The product is available in research preview for ChatGPT Business, Enterprise, Edu, and Teachers plan subscribers. OpenAI describes the feature as “an evolution of GPTs” capable of writing and executing code, connecting to external applications, retaining memory across sessions, and operating without continuous human involvement.
Why It Matters
The launch places OpenAI more directly in competition with Microsoft 365 Copilot and Google Workspace’s Gemini-powered automation features, both of which have made significant inroads with enterprise customers. OpenAI’s Slack integration and scheduling capability specifically address a gap left by session-bound AI tools: multi-step processes that span teams, require asynchronous hand-offs, and run over hours or days rather than within a single conversation. The announcement also signals a strategic move away from the GPT framework that has anchored ChatGPT’s customization layer since November 2023.
Technical Details
Workspace agents are built on Codex and run in the cloud, with each agent provisioned a persistent workspace for files, code, tool integrations, and long-term memory. Per OpenAI’s announcement, agents “can write or run code, use connected apps, remember what they’ve learned, and continue work across multiple steps.” For sensitive operations—such as editing a spreadsheet, sending an email, or adding a calendar event—administrators can configure approval requirements before the agent proceeds. The Compliance API surfaces every agent’s configuration changes and run history for audit purposes. OpenAI also built in defenses against prompt injection attacks that agents may encounter when processing external content during tasks.
Who’s Affected
Organizations on ChatGPT Business, Enterprise, Edu, and Teachers plans gain access immediately through the research preview. OpenAI disclosed three internal production deployments to illustrate scale: a sales agent that qualifies inbound leads and drafts follow-up emails directly in sales representatives’ inboxes; a product feedback agent that monitors Slack, support queues, and public forums to produce weekly prioritized summaries; and an accounting agent that handles month-end close tasks—journal entries, balance sheet reconciliations, and variance analyses—”in minutes” while generating workpapers for reviewer sign-off. Administrators on Enterprise and Edu plans receive additional controls over which user groups can build, share, or invoke specific agents and tool connections.
What’s Next
OpenAI stated that existing GPTs will remain available during the transition period, with a conversion pathway to workspace agents described as “coming soon”—no specific timeline was provided. Additional integration surfaces beyond ChatGPT and Slack are referenced in the announcement as forthcoming, with no platforms named. OpenAI has published templates for finance, sales, marketing, and other categories to give enterprise teams ready-made starting points for their first workspace agents.
