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

# Prompting best practices

> Write effective prompts to get better results from Builder's AI agents.

In the Builder factory, prompts are your briefs to the agents on the line. A clear brief means less rework, fewer credits spent, and software shipped faster. These practices help you get production-quality output on the first pass.

## Start with a clear goal

Before prompting, know what you want:

* **What** — The type of app (dashboard, landing page, tool)
* **Who** — Who will use it
* **Key features** — The 2–3 things it must do in v1

<Note>
  New projects start with the **Product Architect** agent, which helps you define scope before code is written. Use it to plan, then switch to the **Builder agent** to implement.
</Note>

## Be specific

Vague prompts produce vague results. Compare:

| Weak               | Strong                                                                                                          |
| ------------------ | --------------------------------------------------------------------------------------------------------------- |
| "Make a dashboard" | "Build a sales dashboard with a revenue chart, a table of recent orders, and filters for date range and region" |
| "Fix the styling"  | "Make the header sticky, increase button padding to 12px, and use a dark background (#111) with white text"     |
| "Add auth"         | "Add email/password login with a sign-in page, sign-up page, and redirect to /dashboard after login"            |

## Build incrementally

Don't try to build everything in one prompt. Break work into steps:

<Steps>
  <Step title="Layout and navigation">
    "Create the main layout with a sidebar, header, and content area. Add navigation links for Dashboard, Settings, and Profile."
  </Step>

  <Step title="Core feature">
    "Add a task list to the Dashboard page. Each task has a title, status badge, and a checkbox to mark complete."
  </Step>

  <Step title="Polish">
    "Add empty states when there are no tasks. Style the status badges: green for done, yellow for in progress, gray for todo."
  </Step>
</Steps>

<Warning>
  Large, multi-feature prompts often produce incomplete or broken results. Smaller, focused prompts are more reliable.
</Warning>

## Assign the right agent

Each agent is a specialist on your factory floor — use the right one for the stage you're in:

| Agent                 | Factory role                                                    |
| --------------------- | --------------------------------------------------------------- |
| **Product Architect** | Planning stage — PRDs, architecture, scoping before code        |
| **Builder agent**     | Production stage — writes code, implements features, fixes bugs |
| **Task Planner**      | Operations — breaks work into tasks, manages the task board     |
| **Feedback**          | Quality loop — captures bugs, requests, and product signal      |

Switch agents from the agent picker in the chat panel. Starting with Product Architect saves credits by preventing expensive rebuilds.

## Reference what exists

When iterating on an existing project, point to specific parts:

* "Update the **Settings page** to add a dark mode toggle"
* "Change the **primary button color** on the landing page to purple"
* "Add a **new route** at `/analytics` with a bar chart of monthly revenue"

The AI has context about your project's current code and structure.

## Use real content

Replace placeholder text with realistic content early:

* Use actual product names, not "Lorem ipsum"
* Provide sample data that matches your use case
* Include real error messages and edge cases you expect

This helps the AI make better structural decisions.

## Enhance your prompt

Use the **Enhance prompt** button in the chat composer to expand a short idea into a detailed prompt before sending. This is especially useful when you know what you want but aren't sure how to phrase it.

## When something goes wrong

If the AI produces unexpected results:

1. **Be specific about the problem** — "The sidebar overlaps the content on mobile" not "it looks broken"
2. **Ask for one fix at a time** — Don't combine unrelated changes
3. **Revert if needed** — Use the Versions panel to roll back to a working snapshot (paid plans)
4. **Rephrase** — Sometimes rewording the request produces a better result

## Model selection

Different models have different strengths and credit costs:

* **Standard models** — Fast, cost-effective, good for most tasks
* **Premium models** — Better reasoning for complex features (Pro plan and above)

Choose models from the model picker in the chat panel. Premium models consume more credits per message.

## Next steps

<CardGroup cols={2}>
  <Card title="Getting started" icon="rocket" href="/introduction/getting-started">
    Apply these tips to your first project.
  </Card>

  <Card title="Plans & credits" icon="coins" href="/introduction/plans-and-credits">
    Understand model access and credit costs.
  </Card>
</CardGroup>