AI Agent — Generate Work Items from Descriptions

The AI Agent turns a plain-language description into a fully structured set of work items — stories, tasks, and sub-tasks — ready to drop into your sprint. Instead of manually breaking down a feature, you describe what you want to build and the AI handles the decomposition.

Available on R&D boards only.

Accessing the AI Agent

Two ways to open it:

  • Board sidebar → AI Agent icon
  • Command palette (Cmd+K or Ctrl+K) → type "AI Agent"

How It Works

Step 1 — Describe What You're Building

Write a natural language description. You can be as brief or as detailed as you want. The more context you provide, the better the output.

Examples that work well:

A high-level feature description:

"Add a Slack integration so users can get notified when a work item is assigned to them or when a sprint starts."

A bug report:

"Users on mobile can't dismiss the notification panel — the close button is hidden behind the bottom navigation bar on iOS Safari."

A large initiative:

"Build a time-off tracking feature for managers. Managers can set team member availability, mark days as unavailable, and the sprint capacity calculation should automatically exclude those days."

Step 2 — Choose Type

Select whether this is a Feature or a Bug. This shapes how the AI structures the output:

  • Feature: Creates a user story with acceptance criteria, then breaks it into implementation tasks and sub-tasks
  • Bug: Creates a bug report with reproduction steps, then creates a task for investigation and a task for the fix

Step 3 — Generate

Click Generate. The AI analyzes your description, considers your board's existing context, and produces a structured breakdown. This typically takes 5–15 seconds.

Step 4 — Review

Before anything is created, you see a full preview of all generated items:

  • Story or bug report: Title, description, acceptance criteria
  • Tasks: Each with a title, description, and estimated hours
  • Sub-tasks: Attached to the relevant task where applicable
  • Release notes: A draft release note for the story

You can edit any field in the preview — change a title, adjust estimated hours, remove a task that's not needed, or add a task the AI missed.

Step 5 — Create

Click "Create All." Every item is created on your board in one action, with the hierarchy intact (story → tasks → sub-tasks).

Smart Context

The AI Agent doesn't work in isolation. It reads your board's configuration to produce relevant output:

  • Custom fields: If your board has an Epic or Product field, the AI may suggest a value based on the description
  • Existing sprints: If you're in an active sprint context, the AI may suggest assigning generated tasks to it
  • Team members: The AI can suggest assignees based on the area of work described
  • Your board's statuses and stages: Generated items are assigned appropriate starting statuses

Duplicate Detection

When you click Generate, the AI also checks your existing backlog for items similar to the one being described. If potential duplicates are found, they're surfaced in the review step with a similarity score.

You can:

  • Dismiss — the items are genuinely different, proceed with creation
  • Cancel — the description overlaps too much with existing work, abandon creation
  • View existing item — open the existing item to check before deciding

Custom AI Prompts

Board admins can customize how the AI Agent generates items for their specific board.

Board Settings → AI Agent tab

The default prompt works well for most engineering teams. Customize it to:

  • Match your team's task naming conventions ("As a [user], I want..." vs. title-only)
  • Require specific fields to always be populated
  • Set expected task granularity (e.g., "tasks should be completable in 1–4 hours")
  • Add domain context ("this is a fintech product, regulatory compliance tasks should always be included for payment-related features")
  • Set output language if your team works in a language other than English

Changes to the prompt apply to all future generations on that board. The default prompt can be restored at any time.

Best Use Cases

Breaking down large features during planning: Paste in a feature spec or PRD excerpt. Get a sprint-ready task breakdown in under a minute.

Converting bug reports into structured work items: Copy a bug report from Slack, email, or your support tool. The AI formats it correctly, adds reproduction steps, and creates investigation + fix tasks.

Rapid backlog creation: Working on a new product area with an empty backlog? Describe each feature in a sentence or two and generate the initial structure for your entire backlog.

Onboarding new team members: New engineers or PMs who are still learning your task conventions can use the AI Agent to create well-structured items from their descriptions, then refine them over time.

Tips for Better Output

Tip: Include acceptance criteria in your description. "The feature is complete when..." statements dramatically improve the quality of the generated story and help the AI identify the right tasks.

Tip: Mention technical constraints. "This needs to work with our existing webhook system" or "we're using Postgres full-text search, not Elasticsearch" helps the AI generate more accurate implementation tasks.

Tip: If the first generation isn't quite right, don't discard it — edit the preview. It's almost always faster to adjust three fields in the preview than to write the items from scratch.

Related