Workspace agents in ChatGPT: Admin and builder controls

YouTube AI Channels Products

Summary

This video demonstrates the admin and builder controls for ChatGPT workspace agents, enabling team collaboration with governance.

No content available
Original Article
View Cached Full Text

Cached at: 05/23/26, 06:58 AM

**TL;DR:** This video demonstrates the builder and admin control features of ChatGPT workspace agents, ensuring team collaboration efficiency while maintaining security governance. ## Workspace Agent Overview Workspace agents can execute important workflows for your team around the clock. The video uses the "Product Feedback Intelligence Agent" as an example to showcase its core capabilities: - Collect important customer contact information and product feedback from CRM - Create deliverables such as PRD briefings, slides, etc. - Generate actionable tickets and tools like Linear - Provide updates in Slack on a weekly schedule ### Agent Performance in Slack After entering the Slack channel, you can see how the agent integrates contacts from various tools and proactively shares its analysis results in the team's shared Slack channel. It can also respond with additional context, consolidate feedback, and apply that feedback in the next run thanks to its memory. ## Builder Controls Given that the agent can access a wide range of tools and applications, builders have several control options to determine the agent's permissions. ### Read/Write Operation Toggle Builders can toggle specific write and read operations, granting fine-grained control over the agent's access to applications. ### Natural Language Operation Constraints During the build process, builders can also set additional operation constraints using natural language. Example from the video: > "I set an operation constraint so it can only send emails to recipients at openai.com, because the agent is handling sensitive customer and product information." ## Admin Controls In ChatGPT Enterprise, admins govern the entire workspace with powerful role-based access control (RBAC). - Decide who can build and publish agents - Decide which applications and operations agents can access ### App Settings & Permissions Admins can impose additional restrictions on specific apps (e.g., Gmail): - Restrict access to certain apps based on user roles - Constrain parameters and operations for specific apps ### Human Approval Workflow Admins can set up a human approval workflow. For operations with significant impact, ChatGPT prompts the user to confirm, ensuring critical actions undergo human review. ## Summary With workspace agents, teams can move faster, while IT and admins maintain appropriate control. --- **Source:** https://www.youtube.com/watch?v=HaaKUFAOi84

Similar Articles

Introducing workspace agents in ChatGPT

OpenAI Blog

OpenAI introduces workspace agents in ChatGPT, an evolution of GPTs powered by Codex that enables teams to create shared, long-running workflows and handle complex tasks within organizational permissions.

Workspace agents

OpenAI Blog

This article introduces OpenAI's 'Workspace Agents' in ChatGPT, designed to handle repeatable, structured workflows rather than one-off tasks. It explains the core concepts, anatomy, and best practices for using and building these agents for consistent business processes.

Build Hour: Workspace agents in ChatGPT

YouTube AI Channels

OpenAI demonstrated Codex-based Workspace Agents during Build Hours, allowing users to create automated agents that work across systems using natural language in ChatGPT, and showed the complete build process for a meeting preparation assistant.

OpenAI: Workspace Agents for Business

Hacker News Top

OpenAI launches workspace agents for ChatGPT Business and Enterprise plans, allowing teams to build and share AI agents that autonomously handle workflows, integrate with tools, and operate under admin-controlled permissions and audit logs.

Using projects in ChatGPT

OpenAI Blog

OpenAI Academy introduces a guide on using Projects in ChatGPT, dedicated workspaces that consolidate chats, files, and instructions for ongoing work. Features include project-only memory, real-time collaboration, and enterprise-level admin controls.