Workspace agents in ChatGPT: Admin and builder controls

Watch on YouTube

Now Playing

Transcript

67 segments

0:00

Workspace agents can run important

0:02

workflows for your team around the

0:04

clock.

0:05

Let's take a look at this product

0:06

feedback Intel agent which gathers

0:08

important customer contacts and product

0:10

feedback from a CRM,

0:12

creates artifacts such as PRD briefs and

0:14

slides and actionable tickets and tools

0:16

like linear.

0:17

Now, this agent is set up to provide

0:19

these weekly updates in Slack on a

0:21

schedule as you can see here.

0:23

If we go into the Slack channel, you can

0:25

see how the agent supports the product

0:27

team by consolidating contacts across

0:29

the tools they use and proactively

0:31

sharing its analysis in this team's

0:33

shared Slack channel.

0:35

The agent can also respond with more

0:37

context, incorporate feedback, and apply

0:39

that feedback next time it runs due to

0:41

its memory.

0:43

So, given that this agent has access to

0:44

a number of tools and apps, let's take a

0:47

look behind the scenes at the controls

0:48

that agent builders have.

0:51

Agent builders are the ones who

0:52

determine what the agent has access to

0:54

and the actions it's allowed to take.

0:57

They're able to toggle specific write

0:59

and read actions and also set up

1:01

additional action constraints using

1:03

natural language much like the agent

1:05

building process.

1:07

So, here for this agent, I've set up an

1:09

action constraint so that it can only

1:10

send emails to openai.com recipients as

1:13

the agent is handling sensitive customer

1:15

and product information.

1:18

And in ChatGPT Enterprise, admins can

1:20

govern this across the workspace with

1:22

powerful roles-based access controls.

1:25

Admins determine who can build, publish,

1:28

and what apps and actions agents can

1:30

have access to.

1:33

So, let's take a look at the setup for

1:34

apps. I'll quickly run through the

1:36

additional controls an admin has for

1:38

your agent builders and their agents.

1:41

So, if we take a look here at Gmail,

1:43

which was the application we were just

1:44

on, we can take a look at the additional

1:47

admin controls as it relates to

1:49

restricting access to certain

1:50

applications based off a user's role,

1:53

the ability to have further constraints

1:55

on specific app parameters and actions

1:57

as you can see here.

2:00

And last of all, the ability to set up

2:02

human in the loop confirmation flows so

2:04

that chat GPT will prompt for user

2:06

confirmation for actions that it deems

2:07

consequential.

2:10

So all in all, with Workspace Agents,

2:12

teams move faster while IT and admins

2:14

keep the right controls in place.

Interactive Summary

Ask follow-up questions or revisit key timestamps.

The video provides an overview of Workspace Agents, demonstrating how they automate workflows like product feedback analysis, integrate with tools like Slack and CRM systems, and possess memory for continuous improvement. It also highlights the security and governance controls available to agent builders and administrators, including action constraints, role-based access control, and human-in-the-loop verification processes.