Tech • IA • Crypto
Workspace Agents in ChatGPT: Admin and Builder Controls
TL;DR
New workspace AI agents are enabling teams to automate workflows while maintaining strict administrative controls over data access and actions.
KEY POINTS
Automated Product Feedback Analysis
Workspace agents can continuously gather customer contacts and product feedback from CRM systems, transforming raw data into structured outputs such as PRD briefs, presentation slides, and actionable tickets in tools like Linear. This allows product teams to move from insight to execution more quickly without manual aggregation.
Proactive Team Communication
These agents can deliver scheduled updates directly into shared channels such as Slack, consolidating insights from multiple tools into a single, accessible stream. The updates provide ongoing visibility into customer sentiment and product issues, reducing the need for manual reporting.
Context Awareness and Memory
Agents are designed to respond to follow-up questions, incorporate user feedback, and improve over time through built-in memory. This enables more relevant future outputs and creates a feedback loop that refines analysis without repeated manual input.
Granular Control for Builders
Agent builders define what tools and actions an agent can access, including toggling read and write permissions. They can also impose constraints using natural language, such as restricting email outputs to specific domains like openai.com when handling sensitive data.
Enterprise Governance and Access Control
In enterprise environments, administrators manage permissions through role-based access controls. These determine who can build or deploy agents and which applications or integrations are available, ensuring alignment with organizational policies.
Application-Level Restrictions
Admins can further limit agent capabilities within specific apps, such as Gmail, by controlling parameters and allowable actions. This adds an extra layer of security for sensitive workflows involving external communication.
Human-in-the-Loop Safeguards
Systems can require user confirmation before executing consequential actions, introducing a human-in-the-loop mechanism that reduces the risk of unintended outcomes while preserving automation benefits.
CONCLUSION
Workspace AI agents are accelerating operational efficiency while embedding layered controls that allow organizations to balance automation with security and oversight.
Full transcript
Workspace agents can run important workflows for your team around the clock. Let's take a look at this product feedback intel agent which gathers important customer contacts and product feedback from a CRM. Creates artifacts such as PRD briefs and slides and actionable tickets and tools like linear. Now this agent is set up to provide these weekly updates in Slack on a schedule. As you can see here, if we go into the Slack channel, you can see how the agent supports the product team by consolidating context across the tools they use and proactively sharing its analysis in this team's shared Slack channel. The agent can also respond with more context, incorporate feedback, and apply that feedback next time it runs due to its memory. So, given that this agent has access to a number of tools and apps, let's take a look behind the scenes at the controls that agent builders have. Agent builders are the ones who determine what the agent has access to and the actions it's allowed to take. They're able to toggle specific write and read actions and also set up additional action constraints using natural language much like the agent building process. So here for this agent, I've set up an action constraint so that it can only send emails to openai.com recipients as the agent is handling sensitive customer and product information. And in Chat GBT enterprise, admins can govern this across the workspace with powerful rolesbased access controls. Admins determine who can build, publish, and what apps and actions agents can have access to. So let's take a look at the setup for apps. I'll quickly run through the additional controls that an admin has for your agent builders and their agents. So, if we take a look here at Gmail, which was the application we were just on, we can take a look at the additional admin controls as it relates to restricting access to certain applications based off a user's role. The ability to have further constraints on specific app parameters and actions, as you can see here. And last of all, the ability to set up human in the loop confirmation flows so that chat GPT will prompt for user confirmation for actions that it deems consequential. So all in all with workspace agents teams move faster while IT and admins keep the right controls in place.