Skip to main content

What is the task system?

A task is a unit of work that Agent can plan and execute. Each task runs in its own thread — a separate conversation where Agent works independently. You can follow along, give feedback, or let it run in the background. The task system is how Agent breaks work into manageable chunks. When you describe what you want to build, Agent splits your request into discrete tasks and organizes them on a board with columns: Drafts, Active, Ready, and Done.
Task board showing tasks organized in Drafts, Active, Ready, and Done columns
Each task moves from left to right as it progresses, so you can see at a glance what’s planned, what Agent is working on, what’s ready for your review, and what’s finished. Multiple tasks can run at the same time, so you’re not waiting for one to finish before the next one starts. When a task is ready, you review the changes and apply them back to your main version — or dismiss them if they’re not what you wanted.

How parallel tasks work

1

Plan your tasks

Start a conversation in the main thread, or create a new thread and plan there. Agent reviews your project, identifies what would have the most impact, and writes up a detailed plan for each task — including what it will do and what “done” looks like.
Describing what to build in the main thread, with Agent proposing and running tasks
2

Review and accept

Agent proposes a set of tasks — each with a title, description, and a detailed plan you can inspect with View plan. Review the full list, then choose:
  • Accept tasks — Start all tasks and send them to the background
  • Revise plan — Ask Agent to adjust the tasks before starting
Task plans created dialog showing four proposed tasks with Accept tasks and Revise plan options
3

Tasks run in parallel

Once accepted, tasks run simultaneously — each in its own isolated copy of your project. Your main version stays untouched until you choose to apply changes. You can track progress in two ways:
  • Thread view — Each task runs in its own thread with a live status indicator, so you can follow along or jump into any task’s conversation
Thread view showing multiple tasks running in parallel with status indicators
  • Board view — A column-based board showing all tasks organized by status: Drafts, Active, Ready, and Done
Board view showing tasks organized in Drafts, Active, Ready, and Done columns
4

Review and apply

When a task finishes, Agent shows you what it did — the work log, test results, and a live preview of the changes. You decide what happens next:
  • Apply changes to main version — Merge the task’s work into your project
  • Dismiss — Discard the changes if they’re not what you wanted
Agent handles conflict resolution automatically when applying changes from multiple tasks.
Task review screen showing Agent's work log, test results, and Apply changes button

Task states

Every task goes through a series of stages: from an idea to part of your app. Each stage has its own icon so you can tell what’s happening at a glance.
Draft: Agent has proposed this task. It has a plan ready for your review.
Active: Agent is building this task right now. It runs in its own copy of your project so your main version stays safe.
Queued: This task is approved but waiting for a slot to open up before it can start building.
Ready: This task is complete. You can review the changes and apply them to your main version whenever you’re ready.
Applying: The changes from this task are being applied to your main version right now.
Done: This task is finished and its changes are part of your main version.

The task board

All your tasks live on a board organized into columns. Here’s how the board looks as tasks move through each stage.

Drafts and Active

When you describe what you want to build, Agent creates draft tasks — each with a title, description, and a step-by-step plan. You can review the plan, ask Agent to adjust it, or accept it as-is. Once you accept, tasks move to the Active column and Agent starts building them. Each task runs in its own isolated copy of your project, so nothing changes in your main version until you say so. Some tasks depend on others — for example, a task that builds a dashboard needs the database to be set up first. These tasks wait as queued until their dependencies finish. Tasks also queue if you’ve reached your plan’s limit for parallel tasks.

Ready and Done

When a task finishes building, it moves to Ready. This means the work is complete, but nothing has changed in your main version yet — you’re in control. You can apply changes in two ways:
  • From the thread — Open the task’s thread and you’ll see a review prompt like the one below. Press Apply changes to main version to bring the work into your project, or Dismiss to discard it.
  • From the board — Click the three-dot menu on any Ready task card to see your options, then press Apply changes
Once applied, the task card moves to the Done column. Agent handles conflicts automatically when you apply changes from multiple tasks. If something doesn’t look right, you can always ask Agent to fix it.

What you can build

The task system works for any project. Here are a few patterns where it shines:
  • Ship faster with parallel tasks — Build authentication, a dashboard, and an API layer simultaneously instead of sequentially
  • Plan complex projects — Break a large feature into well-scoped tasks with dependencies, then execute with confidence
  • Explore design directions in parallel — Try multiple UI approaches at the same time and pick the best result

Frequently asked questions

No. The main thread works exactly the same. Agent automatically creates tasks behind the scenes when it helps. You can also use Plan mode to see and manage tasks explicitly.
Parallel execution supports up to 10 concurrent tasks on Pro and up to 2 on Core (for a limited time). If you’re hitting the limit, additional tasks queue and run as slots free up.
Agent uses AI-assisted conflict resolution. It understands what each task was trying to accomplish and applies the changes intelligently. If something looks off, you can ask Agent to fix it in the thread.
Yes. Agent automatically detects dependencies — for example, a task that builds a dashboard depends on the task that creates the database schema. Dependent tasks wait until their prerequisites complete.

Availability

CapabilityCorePro
Task planning
Background tasks
Parallel executionUp to 2*Up to 10
* Available on Core for a limited time.