> ## Documentation Index
> Fetch the complete documentation index at: https://docs.pre.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Build Modes

> Control how autonomous agents work through your project.

Once your specification is ready, you choose how agents execute the work. pre.dev offers three build modes that give you different levels of control.

```
  Build Next Task
> Build On Autopilot
  Chat
```

<Tabs>
  <Tab title="Build Next Task">
    Agents build **one task at a time**, then wait for your review before proceeding.

    <Steps>
      <Step title="Task Assigned">
        The next task in the spec is assigned to a build agent
      </Step>

      <Step title="Agent Builds">
        The agent implements the task in a secure sandbox
      </Step>

      <Step title="Verification">
        Acceptance criteria are verified (types, lint, tests, browser)
      </Step>

      <Step title="PR Opened">
        A pull request is opened for your review
      </Step>

      <Step title="You Review">
        Review, request changes, or approve — then click "Build Next Task" to continue
      </Step>
    </Steps>

    <Tip>
      **Best for:** Learning how pre.dev works, sensitive codebases, and teaching the agent your preferences by reviewing early tasks.
    </Tip>
  </Tab>

  <Tab title="Build on Autopilot">
    Agents work through the **entire roadmap autonomously**. All tasks execute in sequence (or parallel where dependencies allow), and PRs are opened for each.

    <Steps>
      <Step title="Click Build All">
        Agents pick up tasks following the milestone order
      </Step>

      <Step title="Parallel Execution">
        Independent tasks within the same milestone run in parallel
      </Step>

      <Step title="Continuous Verification">
        Each task goes through acceptance verification automatically
      </Step>

      <Step title="PRs Accumulate">
        Review and merge at your pace — the roadmap shows real-time progress
      </Step>
    </Steps>

    <Tip>
      **Best for:** Projects where you trust the spec and want maximum speed, overnight builds, or well-defined projects with clear requirements.
    </Tip>
  </Tab>

  <Tab title="Chat">
    An interactive mode where you **converse with the build agent** directly. Ask questions, give instructions, or guide implementation manually.

    <Steps>
      <Step title="Open Chat">
        Open the Chat interface for your project
      </Step>

      <Step title="Instruct the Agent">
        Ask anything — about the spec, the codebase, or give specific build instructions
      </Step>

      <Step title="Agent Implements">
        The agent implements your request with full project context
      </Step>

      <Step title="PR Created">
        Changes appear as PRs just like the other modes
      </Step>
    </Steps>

    <Tip>
      **Best for:** One-off changes, debugging, asking the agent to explain code, or quick fixes that don't need a full spec.
    </Tip>
  </Tab>
</Tabs>

## Choosing a Mode

| Scenario                 | Recommended Mode                           |
| ------------------------ | ------------------------------------------ |
| First time using pre.dev | Build Next Task                            |
| Small MVP you want fast  | Build on Autopilot                         |
| Complex production app   | Build Next Task (at least for Milestone 1) |
| Quick feature addition   | Chat                                       |
| Overnight build          | Build on Autopilot                         |
| Exploratory / learning   | Chat                                       |

## Mixing Modes

You can switch between modes at any time:

* Start with **Build Next Task** to validate the first few PRs
* Switch to **Build on Autopilot** once you're confident the agent understands the patterns
* Drop into **Chat** anytime to make a quick adjustment or ask a question

## What Happens During a Build

Regardless of mode, each task goes through the same pipeline:

```
Task → Agent codes → Verification → PR opened
                          |
                     Failed? Auto-fix
                     (loops until pass)
```

If acceptance verification fails (types don't compile, tests fail, linting errors), the agent automatically iterates until the checks pass. You never see broken code in a PR.