All articles

AI Agent Actions Reference

Last updated on

Wrike Documentation Team

Wrike Documentation Team

TL;DR

Wrike AI agents perform actions when triggered - post comments, update fields, change assignees/status/dates, move items, send emails, or notify via Slack/Teams. You can add multiple actions per agent (run in order), filter which items they affect, and set target locations.

Table 10. Availability

Availability: Business, Pinnacle, Apex. ; Unavailability: Free, Team;

Overview

Actions define what an AI agent does when triggered. You can add multiple actions per agent.

Actions Table

Action What it does
Post comment Share analysis or decision as a comment on the relevant item.
Change custom field Updates specific custom fields (choose which to update).
Change assignee Assign work items to users based on your prompt.
Change work item name Updates the task or project title based on analysis.
Change status Updates the status based on conditions you define (standard & custom workflows).
Select Link-to-Database record Changes which database record is selected in L2DB fields.
Change start/due date Modifies start and due dates on tasks.
Location (Move / Add) Moves or adds work items to folders defined via location chips.
Start or manage approval flow Creates new approvals or updates existing draft approvals.
Send email (Gmail/Outlook) Sends an email via a connected OAuth account.
Notify on Slack Posts a message to a connected Slack channel.
Notify on Microsoft Teams Posts a message to a connected Teams channel.

Custom Field Write Rules

AI agents can update these custom field types: Text, Single Select, Multi Select, Number, Percent, Currency, Link to Database, Checkbox, Date, and Duration.

  • Read-only (cannot be written): Mirror fields and Calculated/Formula fields (readable as context). People-type custom fields cannot be read or written at all.
  • Date fields that include a time component (Wrike Labs feature) are not supported in AI agent actions and won't appear in the field picker. If a date field is changed to include a time part after an agent was saved with that field, the action will fail on every execution.

Action Details & Prompt Examples

Change assignee - AI agents can see user profile attributes (job title, department, country, timezone, assigned task count) and group membership. Agents cannot access workload or capacity data beyond task count. (See Assign Work Items with AI Agents.)

Select Link-to-Database record:

  • Change L2DB Field action: AI agent can search the entire database. Search by value, not column name.
  • Other actions: AI agent sees only the currently selected value.

✅ "select the record containing the value" 
❌ "select the record where Column Name is value"

Change start/due date - example prompts:

  • "When status changes to Blocked, push the due date forward by 5 working days."
  • "When all required fields are filled, set the start date to today."

Location - two modes:

  • Move to location: Removes item from current folder, places it exclusively in target.
  • Add to location: Keeps item in current folder and adds it to an additional folder.

Example prompts:

  • "When a task description mentions legal, compliance, or regulatory, move it to the Legal Review folder."
  • "When Priority is set to Critical, add the task to the Urgent Queue folder."

Important: AI agents can only move items to folders you explicitly reference as location chips. They cannot move items to arbitrary locations or create new folders.

Start or manage approval flow - supported actions: create approval, update draft approval, add/remove approvers, choose author or assignee as approver, specify users/groups via user chips.

Example prompts:

  • "When status changes to Ready for Review, start an approval with the project owner as approver."
  • "When the Budget field exceeds 50,000, start an approval with the Finance Team user group."

Send email - During setup, the space admin authenticates a Gmail or Outlook account.

The AI agent can derive addresses from: assignee's primary email, author's email, custom field values, task title/description text, explicit addresses, composition rules (e.g., "first name + dot + last name + @contoso.com").

The AI agent cannot get addresses from: followers, secondary email addresses.

Tip: Use unambiguous language like "Send an email to the assignees summarizing the project status" rather than "Update assignees on the project status."

Important: A single email action can address multiple "To" recipients. No CC/BCC, no attachments, no secondary emails, no follower emails. Sender shows the connected account.

Notify on Slack - A Space admin connects a Slack workspace and selects a channel via a chip; the channel is fixed at setup time.

Important: Requires connecting a Slack workspace. The Slack app must be installed in any target channel. The Wrike app for Slack is Marketplace-approved; restricted workspaces may need admin pre-approval.

Limitations: one channel per action; no thread replies; no @mentioning Slack users.

Notify on Microsoft Teams - A Space admin connects a Teams account and selects a channel; fixed at setup time. Messages include a link back to the work item; activity log records delivery status.

Limitations: channels within a team only (no group DMs); no @mentioning; no thread messages; one channel per action.

Action Target Location

Choose where the update should happen:

Target Scope
Work item where the AI agent was added The parent item where the agent is configured.
All subitems of the work item where the AI agent was added All children (max 1,000 subitems per execution).
Work item where the trigger happened The specific item that triggered the agent.
Subitems of the triggered item Direct children only (one level below the triggered item).

Work Items Filtering

Each action has a filter picker - filter by item type, status, assignee, custom fields, importance, name, or any combination. Items that don't match are skipped without consuming agent credits. Configured per action.

Tip: If your AI agent is appointed to a folder with both projects and tasks, use a filter to restrict actions to the item types you care about.

Multiple Actions

How multiple actions work:

  • Actions are independent - one action's result does not affect others.
  • Guaranteed execution order - Actions now run in the exact order you configure. Since execution is sequential, Action B can read what Action A wrote. Simply drag and drop actions to set their order in the builder.
  • Each action has its own instruction and target; the General instruction is shared.
  • Name each action for easier identification in config and logs (e.g., "Set Priority", "Route to Team", "Post Summary").

What this means for your prompts:

  • Every action prompt must stand alone.
  • Don't write "Read the comment posted by the previous action" or "If Action 1 succeeded…".
  • If two actions need the same value, have each calculate it independently from the same source.
  • If Action B genuinely needs Action A's output → use agent chaining (see Write Effective Agent Prompts).

Good multi-action use cases: set a category field AND post a comment; change assignee AND update status; update multiple custom fields simultaneously.

When to Use Status Change

  • Intake validation: Change to Ready when all required fields complete.
  • Overdue tracking: Mark Overdue when past due date.
  • Dependency detection: Set Blocked when dependencies are mentioned.
  • Project completion: Move to Complete when all subtasks finished.

Notes: Agents respect workflow transition rules and cannot skip required statuses. Changes are logged with reasoning. If status is already correct, only a log entry is created. Works with standard and custom workflows.

Example status prompts:

  • "When a task passes its due date and status is not Completed, change status to Overdue."
  • "When a task has all three required fields populated

What’s Next?

0 comments

Please sign in to leave a comment.

Top