OpenAI is rolling out workspace agents for ChatGPT, a significant expansion aimed at transforming team collaboration and automating complex workflows. These agents, built upon the Codex model, are designed to handle multi-step tasks and long-running processes, operating within the security and governance parameters set by organizations.
This evolution of GPTs allows teams to create shared AI assistants that can perform a variety of work-related functions, from report generation and code writing to responding to internal communications. Running in the cloud, these agents can continue tasks even when users are offline, facilitating continuous progress.
Automating Workflows Across Teams
Workspace agents bridge the gap for workflows that require shared context, handoffs, and cross-team decision-making. They can access relevant systems, adhere to established team processes, and even request human approval for critical steps, ensuring work remains on track across different tools.
OpenAI’s own sales team is reportedly using an agent to streamline lead qualification, pulling data from call notes and research to draft follow-up emails directly into sales representatives' inboxes. This aims to free up account teams for more customer-facing activities.
These capabilities are currently in research preview for ChatGPT Business, Enterprise, Edu, and Teachers plans. Users can initiate agent creation by describing a workflow, with ChatGPT assisting in defining steps, connecting tools, and testing functionality.
Building and Deploying Agents
The platform offers examples of potential agents, including a software review agent for triaging requests, a product feedback router, a weekly metrics reporter, a lead outreach agent, and a third-party risk management agent. OpenAI also provides templates for common business functions.
Agents can write and run code, utilize connected applications, and maintain memory across interactions. They can be scheduled or deployed within platforms like Slack, enabling them to respond to incoming requests in real-time.
Teams can manage and discover shared agents through a dedicated tab in the ChatGPT interface. This allows for the codification of best practices into reusable, organizational workflows.
For example, an accounting team built an agent to automate parts of the month-end close process, generating workpapers and ensuring adherence to internal policies.
Control and Governance
OpenAI emphasizes user control, allowing administrators to define what tools and data agents can access, and what actions they can perform. Sensitive operations can be configured to require human approval before execution.
Enterprise-grade controls are integrated for administrators to manage access to tools, actions, and agent building/sharing capabilities. Built-in safeguards are designed to protect against misleading external content and prompt injection attacks.
The Compliance API offers administrators visibility into agent configurations, updates, and usage, enabling monitoring and control. Future updates will provide a centralized view of all organization-wide agents within the admin console.
Early feedback from companies like Rippling, SoftBank Corp., Better Mortgage, BBVA, and Hibob highlights the potential for increased consistency and time savings, enabling employees to focus on higher-value work.
Workspace agents are available in research preview until May 6, 2026, after which credit-based pricing will be implemented. OpenAI plans to introduce new triggers, performance dashboards, expanded tool integrations, and support for agents within its Codex application.
This initiative represents a step toward AI that operates within the flow of work, aiming to reduce coordination overhead and boost team productivity. For more details, explore workspace agents in ChatGPT, as announced by OpenAI News.
