Use Work Agents

The highlighted information on this page refers to functionality not yet generally available. It is available only in the Preview environment for all customers. After the release to Preview, the same features are also available monthly in the Production environment for customers who enabled fast releases.

For information about fast releases, see Enable or disable fast releases for your organization.

Work Agents are AI Collaborators that can be assigned directly to Workfront tasks, issues, and requests. Like other AI Collaborators, Work Agents are configured in the Setup area and assigned to work items just like a user.

Work Agents connect to agents that you have configured in Copilot Studio, Claude, OpenAI or IBM.

IMPORTANT
Writer has deprecated their use of agents. Work Agents configured using Writer agents no longer work after October 9. 2026.
For information on the deprecation, see Agent Library migration & deprecation in the Writer documentation.

For information and instructions about creating a Work Agent in Workfront, see Configure a Work Agent in the article Configure AI Collaborators.

Access requirements

Expand to view access requirements for the functionality in this article.
table 0-row-2 1-row-2 2-row-2 layout-auto html-authored no-header
Adobe Workfront package Select, Prime, or Ultimate
Adobe Workfront license Standard
Access level configurations System Administrator

For information, see Access requirements in Workfront documentation.

Prerequisites

  • You must configure an agent in Copilot, Claude, OpenAI, or IBM before you can use it as a Work Agent in Workfront.

Work Agent overview

Work Agents are a way to assign agents to specific tasks in Workfront. You configure the agent in an app such as Copilot Studio or Claude.ai, then connect that agent to Workfront as a Work Agent. You can then assign it to tasks as you would assign a user.

Some example workflows may include:

  • Generating copy from task details and attached text-based documents, then posting the output to the update stream or uploading it as a text-based document.
  • Generating copy from a task description, reviewing the copy against guidelines configured in the agent, and posting copy to the update stream.
  • Reading details of an event, identifying missing details, and posting questions in the update stream about the missing details.
NOTE
  • Specific details about an agent’s responsibilities and abilities are configured in the application where the agent is created, not in Workfront.
  • The Workfront MCP server does not need to be added to the agent used as a Work Agent, and does not need to be connected for the Work Agent to work.
  • Work Agents currently support agents created in Copilot Studio, Claude, OpenAI, and IBM.
  • When configuring an agent in Copilot Studio, you must set security to No authentication.
  • For information and instructions about creating a Work Agent in Workfront, see Configure a Work Agent in the article Configure AI Collaborators.

Information a Work Agent reads

When a Work Agent begins work on a task, it automatically reads the following task information as context:

  • Task title
  • Task description
  • Comments in the task’s update stream
  • Information in any custom form attached to the task
  • Attached text-based documents, such as DOC and PDF files
  • Project fields, if they are set up in the Work Agent configuration

This information is always read and is not configurable as a Workfront setting.

TIP
For best results, we recommend:
  • Including any background information you want the agent to use directly in the task description or a relevant custom form field.
  • Making sure the task matches what your agent is instructed to do. For example, if your agent is instructed to translate text from English to French, include the text you want translated in the task description.

Actions a Work Agent can take

Depending on how the Work Agent is configured in Workfront, it can take the following actions after it completes its work:

  • Write task fields, if configured
  • Upload text-based documents
  • Mark the task complete

Work Agent start triggers

When a Work Agent is assigned to a task or issue, it begins work when any of the following situations are met:

  • The Work Agent is assigned to a task that is ready to start. (For example, if the task has predecessors, the predecessors are complete.)
  • The Work Agent and a user are assigned to a task, and the Work Agent is assigned first.
  • A task to which a Work Agent is already assigned as becomes ready to start, and the Work Agent is the only or primary assignee. (For example, if the task has predecessors, the predecessors are complete.)
  • A task to which a Work Agent and a user are already assigned becomes ready to start, and the Work Agent was assigned first or is the primary assignee. (For example, if the task has predecessors, the predecessors are complete.)
  • A user and a Work Agent are assigned to a task, and the user is removed.
  • A user and a Work Agent are assigned to a task, and the Work Agent is set as the Primary Assignee for the task.

The following situations do not cause the Work Agent to begin work on the task:

  • A Work Agent is assigned to a task that already has a user assigned.
  • A Work Agent is @mentioned in a task.
  • A Work Agent is assigned to a task that already has a Work Agent assigned. In this case, the first Work Agent assigned will have already begun the work, and the second Work Agent will do nothing.
  • A Work Agent is assigned to a task that is not ready to start. (For example, if the task has predecessors, the predecessors are not yet complete.)

Assign a Work Agent to a task, issue, or request

Work Agents are assigned to tasks, issues, or requests the same way users are assigned.

When you are searching for a Work Agent in the list of available assignees, the name of the Work Agent is a first name only.

For instructions, see Assign tasks and Manage work and team requests.

NOTE
Work Agents cannot be assigned to review or approve a document.

Troubleshooting Work Agents

If your Work Agent does not return a response or output, please check the following:

  • Make sure your agent is published on the AI platform provider side.
  • Make sure you have sufficient AI credits with your agent’s platform.
  • Make sure the the action taken on the task does not require a specific access level.
  • If you are using Copilot as the agent provider, ensure you are using the “no authentication” setting.
  • If you are using Copilot, make sure that your agent is configured on a global environment. Work Agent functionality does not currently support regional versions of Copilot Studio.
  • Make sure that the Collaborator is the primary assignee on the task.
  • Make sure that the task that the Work Agent is assigned to Can Start. For example, check to see that all task predecessors to that task are complete.
TIP
You can also go to the agent provider platform and ask the agent to perform the task within the platform. If the agent cannot perform the task within the platform, then the Work Agent will also run into issues in Workfront.
recommendation-more-help
workfront-help-quicksilver