OpenAI officially launched “Workspace Agents” in research preview on Wednesday, April 22, 2026. Designed as a significant evolution of the “GPTs” framework, these agents are autonomous, persistent assistants that can handle complex, multi-step workflows for teams even when the user is offline.
Unlike traditional chatbots, Workspace Agents run in the cloud, possess long-term memory, and can orchestrate actions across multiple connected third-party tools like Slack, Google Drive, and CRM systems.
1. Key Capabilities & Workflow Examples
OpenAI is positioning these agents as “digital employees” that can own recurring business processes.
- Autonomous Execution: Agents can run on a schedule or be triggered by specific events (e.g., an incoming email or a Slack message).
- Multi-Tool Integration: They can gather context from one system (like a call recording), process it, and take action in another (like drafting a follow-up email in a CRM).
- Cloud Persistence: Because they run in the cloud, they don’t require the ChatGPT app to be open to finish a task.
Example Use Cases:
| Agent Type | Functionality |
| Product Feedback Router | Monitors Slack and public forums; creates prioritized tickets in Jira. |
| Software Reviewer | Checks employee software requests against company policy and files IT tickets. |
| Weekly Metrics Reporter | Pulls data from databases, creates charts, and shares a summary report. |
| Sales Outreach Agent | Researches leads, scores them, and drafts personalized follow-up emails. |
2. Technical Foundation: Powered by Codex
The agents are powered by Codex, which allows them to write and execute code in real-time to solve problems. This makes them significantly more capable than standard GPTs at handling data analysis and technical troubleshooting.
- Memory & Correction: Agents learn from user feedback over time, improving their accuracy and adherence to specific organizational processes.
- Shared Logic: Teams can build an agent once and share it across the organization, allowing multiple departments to use the same automated logic.
3. Availability and Pricing
OpenAI is rolling this out as a “Research Preview” with a unique introductory pricing structure.
- Eligible Plans: Available for ChatGPT Business, Enterprise, Edu, and Teachers subscribers.
- Introductory Offer: The feature is free to use until May 6, 2026.
- Credit-Based Model: Starting May 7, 2026, usage will shift to a credit-based pricing model, where organizations pay for the “computational work” the agents perform.
- Transition Path: OpenAI has confirmed that GPTs are not being discontinued; a mechanism to “convert GPTs into Workspace Agents” will be released later this year.
4. Security and Admin Controls
To address enterprise concerns about autonomous AI, OpenAI has included robust governance tools:
- Role-Based Access: Admins can control who can build, share, or trigger specific agents.
- Human-in-the-Loop: For sensitive actions (like sending emails or editing financial spreadsheets), agents can be configured to require human approval before proceeding.
- Prompt Injection Protection: Built-in safeguards are designed to prevent agents from being manipulated by misleading external data.
