Skip to main content
Handling multiple pieces of work at once is one of those skills that separates good builders from exceptional ones. This page teaches you how to build several features at the same time, safely, even if you’ve never worked this way before.
Café Opening Site task board showing one independent task running while two tasks wait for a prerequisite

​
A new way to work with Agent

So far, you’ve probably worked with Agent like a partner sitting next to you: you ask, you watch, you respond. Building in parallel is a different relationship. You hand a feature to Agent as a task, and it goes off and builds it in the background while you do something else. You already work this way in everyday life. You don’t stand in front of the washing machine until it finishes. You start the laundry, then go cook. Tasks work the same way: start one, and while it runs, start another, review something else, or just keep chatting with Agent in the main conversation. This changes your role. You stop being a passenger watching one build and become the director of several.

​
Your app is safe while tasks run

The first question everyone asks: will a background task break my app while I’m not looking? No. Each task works on its own copy of your project. Your app doesn’t change until you review a task’s work and choose to apply it. Each task also gets its own Agent and inherits your full project context, so you don’t repeat yourself. Tasks are independent: you can start the next one the moment the first is running.

​
Build in parallel

1

Open your project and pin the sidebar

Start from the default Build layout: Agent chat on the left and Preview on the right. Expand the global sidebar and pin it so the project’s task controls and status updates remain visible.
2

Create a task from the project

Expand the project in the sidebar, then select the + button beside its name. This is the quickest way to start a background task.
Another way to create a task is to hover over the board icon beside the project and select New task. From the same menu, you can open the board and use Plan a new task… at the bottom.
Project board shortcut menu with Board, Open, and New task actions
3

Describe one focused outcome

Give the task one clear result, then send it. Agent inspects the project and proposes a plan without changing the main version.
4

Review and approve the plan

Check the scope, implementation steps, definition of done, relevant files, and agent mode. Select Start building only when the plan matches your intent.
A proposed background-task plan awaiting Start building approval
After you select Start building, the task begins working in the background while you keep your place in the main version.
An approved task starting to build, with the task conversation beside the Preview
5

Create and approve the remaining tasks

Repeat the process one task at a time. Give independent work separate outcomes. Describe dependent work in terms of the feature or project structure it requires so Agent can recognize the prerequisite.When a new task needs something another task is still building, Agent names the prerequisite in its plan and schedules the new task to begin after that work is complete and approved.
Agent's plan naming a prerequisite task and scheduling the new task to start after it completes
6

Let Agent schedule the work

Open the board to follow Drafts, Active, Ready, and Done. Independent tasks run when capacity is available. Dependent tasks show what blocks them and remain queued until the prerequisite is applied.
7

Review and apply ready work

Open each Ready task, inspect its work and test results, and try the feature in Preview. Select Apply when it meets your expectations. Applying a prerequisite automatically unlocks the tasks that depend on it.
Task board with a completed prerequisite in Ready and visible Apply and Review actions
After you apply the prerequisite, its dependent tasks move into Active automatically.
Task board after applying task one, with its dependent tasks automatically active
8

Verify the main version

Return to the default Build view after every apply. Use Preview to try the result in the main version, and consult the Agent chat for the implementation summary, checks, warnings, or anything that still needs attention.
Default Build view with Agent chat beside the main café site Preview
Repeat the review, apply, and verification loop until every approved task is in Done.
Task board with all four approved tasks in Done

​
A worked example: four features at once

Imagine a café opening site. You want to add four focused improvements:
  • An accessible skip link and stable landmark IDs
  • Compact section navigation that uses those IDs
  • Better newsletter form feedback that also builds on the accessibility foundation
  • Open Graph metadata, which only changes the document head
The interesting part is deciding what depends on what. The metadata task only touches the document head, so it shares nothing with the other three and can run immediately. The navigation and newsletter tasks both build on the landmark task’s stable IDs and page structure, so they must wait for it — describe them in terms of that structure and Agent recognizes the prerequisite on its own. You don’t have to track that schedule yourself: blocked cards name their prerequisite on the board, while unrelated work runs as Active.

​
Slice your work small

The skill that makes parallel building work is cutting features into small, self-contained tasks. Small tasks finish faster, are easier to review, and are less likely to interfere with each other. A useful rule: if your task description contains “and”, consider splitting it.
  • Good: “Add an events page”, “Add a contact form”
  • Too big: “Add events and registration and email notifications and an admin view”
The big version isn’t wrong, but it forces one Agent to do everything in sequence. Four small tasks give Agent the chance to run them together. Test yourself. Which of these could run in parallel?
  1. Add a dark mode toggle
  2. Add a customer reviews section
  3. Let customers reply to reviews
  4. Translate the app into French
Tasks 1, 2, and 4 are independent and can all run at once. Task 3 depends on task 2: there’s nothing to reply to until reviews exist. You don’t need to catch this yourself. Agent detects the dependency and sequences those two tasks for you.

​
Apply changes safely

Each task works in an isolated copy of your project. When you apply one task, your main version changes while other tasks may still be based on an earlier version.
Agent handles conflicts automatically when you apply changes from multiple tasks. Review the resulting app after each apply. If something looks wrong, ask Agent to fix it in the task thread or main conversation.
When two tasks change the same file, the Ready card can temporarily show Resolving conflicts after you select Apply. Wait for Agent to finish merging and rechecking the result; you do not need to resolve the files manually.
Task board showing Agent resolving conflicts while applying a task

​
Review like a manager

Parallel building makes you a manager, and the classic manager mistake is approving work you didn’t check. The discipline is a simple loop, one task at a time:
1

Test the task's work

Open the finished task and try the feature. Click through it like a user would.
2

Apply it

Happy with it? Apply the changes to your main app.
3

Check the main version

Confirm the applied feature works with the changes already in your app.
4

Repeat

Test and apply the next ready task until every task has landed.
Never batch-approve. Three unreviewed features applied at once means three suspects when something looks wrong.

​
When to stay sequential

Parallel building is a tool, not a rule. Stay in the main conversation, one thing at a time, when:
  • Two ideas touch the same screen. “Redesign the home page” and “change the navigation” will collide. Do them in order.
  • You’re exploring. In design work, the result of one attempt changes what you want next. Exploration is a conversation, not a work order.
  • You can’t describe the task in one sentence. If you can’t state it simply, you’re not ready to delegate it.

​
Work within your limits

Your plan controls how many background tasks can run at once. Additional tasks queue and start as slots become available. See Task system for the current limits. Your attention is a limit, too. Start with two tasks. If reviewing them feels comfortable, try three. You don’t need to supervise every step, but you still need to test each result before applying it.

​
Next steps