Using AI Coworker

AI Coworkers are agents you define to help you get more work done, and are available to all Groupthink users at https://agendas.groupthink.com/assistants

Groupthink currently offers a general-purpose AI Coworker called an Assistant.

Each assistant has a name, username, job title, and job description. Assistants can be given responsibilities, which are tasks they should perform based on things you'd like them to monitor, such as your email, GitHub, or meetings you're attending or invited to.

Assistants belong to the Organization they're created in, and anyone in the Organization may create or modify an Assistant.

Responsibilities

Responsibilities define work that an assistant should do.

Each Responsibility has a trigger that defines when the responsibility should be performed, tasks that define the work that should be done, and an option to define extra information that may help the assistant perform the task better.

Responsibilities may be enabled or disabled, allowing you to control whether the Assistant should perform the work.

Connections

Assistants know when to run Responsibilities by watching for changes in the data sources you've given them access to. For example, if you've given an Assistant access to your GitHub organization, it will know when a new issue is created and can run a Responsibility based on that trigger.

Authorizing an Assistant to access an external system is done by allowing the Assistant to access a Connection you've created in Groupthink. Different Connections offer different methods of access, and you may define this when creating the Connection. Keep in mind that in many cases, the Assistant may be acting as you when utilizing the Connection.

Monitors

Once an Assistant has been authorized to use a Connection, you may define one or more Monitors that instruct the Assistant to "watch" for changes in the data source. For example, you may create a Monitor that watches for new emails in your inbox, or new issues in a GitHub repository.

Keep in mind that you may want to create multiple Monitors for the same Connection or data source. For example, you may want your Assistant to watch for changes in two different GitHub organizations, or two different email accounts.

Chatting with an Assistant

After you create a new Assistant, you will be added to a new chat room with the Assistant. You may ask the Assistant what work they can perform, and they will answer based on the Connections they have access to and the Responsibilities they have been given.

The Room will only be you and the Assistant to start, but you can add anyone else in your Organization by clicking "Share" in the Room.

Using AI Coworkers

Creating an Assistant

Assistants are AI-powered systems that can do various tasks for you.

You'll provide your assistant with responsibilities it should perform based on things you'd like it to monitor, such as your email, GitHub, or meetings you're attending or invited to.

Assistants belong to the Organization they're created in.

  1. Click "AI Team" in the left hand navigation bar
  2. Click "Create Assistant" button in the upper right
  3. Give your assistant a name and username (so they can be mentioned in chats and docs)
  4. Describe what your assistant does
  5. Click "Create Assistant"
  6. Your assistant will be added to the list of AI Coworkers

Managing an Assistant

  1. Click "Manage [Assistant Name]"
  2. Give your assistant some responsibilities by clicking "New Responsibility"

Creating Responsibilities

Responsibilities define work that an assistant should do.

They are made up of a trigger, which is the event that causes the work to happen, and a set of instructions that the assistant should follow.

Trigger

Explain when work should happen.

For example:

  • "A new email is received"
  • "A new Github issue is added to the API repo"
  • "You're mentioned in a chat and asked to create copy for a social media post"
Instructions

Explain what work should happen.

This is a free-form text field, so you can just write instructions without knowing anything about how they will be implemented. However, the more structure and opinion you give the better the assistant will perform.

You'll be able to iterate on this later after you see how your assistant performs.

For example:

  • Consider using a numbered or bulleted list
  • Be very explicit about the output you want, with as much detail as possible
  • Be very specific about the output you want, with as much nuance as possible
Additional Information

Add any important details or contextual information here. This could include details about the trigger, the tasks, how this fits into the overall work you and your team is doing, and even encouragement for your assistant.

The more the assistant knows, the better it can perform. This is also a great way for you to capture the why behind what you're asking them to do.

Configure Monitors

TKTKTK

Frequently Asked Questions

How Many AI Coworkers Can I have?

AI Coworkers is a recently launched feature, and for now you can create unlimited coworkers. This may change in the future.