> ## Documentation Index
> Fetch the complete documentation index at: https://vlyai-1c28d863.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Iterating effectively

> Patterns for making 50 small changes a day in vly without losing momentum. The mindset, the rhythm, and the safety nets.

vly's loop is fast — 30 seconds from prompt to deploy. The temptation is to fire off prompts as fast as you think them. The teams that ship fastest aren't the ones with the most prompts; they're the ones whose prompts land cleanly the first time.

## The rhythm

<Steps>
  <Step title="Look at the live app">
    Don't prompt from memory. Open the preview, see the current state, decide what's next.
  </Step>

  <Step title="One change at a time">
    Five small prompts beat one big one. Easier to roll back, easier to reason about, less likely the agent misreads.
  </Step>

  <Step title="Use Plan mode for anything that changes the schema or adds a page">
    The 5-second review prevents 5-minute redos.
  </Step>

  <Step title="Verify in the preview">
    Don't trust "Build complete." Click around. If you see something off, fix before moving on — every layer you build on top of a bug compounds.
  </Step>

  <Step title="Commit the revision when satisfied">
    Each prompt creates a revision; if you regret it, roll back. No commitment penalty for trying something.
  </Step>
</Steps>

## When you're stuck on a prompt

<CardGroup cols={2}>
  <Card title="Read the agent's first response" icon="eye">
    If it misses the point, the prompt is the problem. Roll back, refine, re-prompt — don't fight the wrong implementation.
  </Card>

  <Card title="Add a constraint" icon="brackets">
    "Same change, but don't break X." Often a prompt produces correct primary behavior + unexpected side effects. Constraints fix the side effects.
  </Card>

  <Card title="Switch to Plan mode" icon="workflow">
    If you've re-prompted twice and the result still misses, drop into Plan mode for the next attempt. The plan reveals the misunderstanding.
  </Card>

  <Card title="Switch agents" icon="bot">
    Different agents have different strengths. If Claude Code keeps over-editing on a tight change, try Codex.
  </Card>
</CardGroup>

## Avoid these patterns

<AccordionGroup>
  <Accordion title="Bundling unrelated changes" icon="package">
    "Add a settings page, fix the broken date picker, and refactor the auth flow." Three prompts, not one. If any single piece fails, the whole revision is suspect.
  </Accordion>

  <Accordion title="Prompting from memory" icon="brain-cog">
    "Add a button next to the avatar." Open the live app — is there an avatar? Is it where you remember? Memory fades fast in a 50-revision session.
  </Accordion>

  <Accordion title="Skipping the verify step" icon="eye-off">
    "Build complete" doesn't mean correct. Click around the preview. Verify the change is what you wanted, *and* that it didn't break anything else.
  </Accordion>

  <Accordion title="Big prompts when small ones work" icon="maximize">
    "Add billing, subscriptions, plans, the billing portal, dunning, and tax." Five prompts in sequence beats one prompt with five things.
  </Accordion>

  <Accordion title="Ignoring custom instructions" icon="x-circle">
    If you keep correcting the same thing ("use named functions, not arrows"), put it in [custom instructions](/building/custom-instructions). One-time setup; works on every future prompt.
  </Accordion>
</AccordionGroup>

## The "lock and step" pattern

For a complex feature, work in two phases:

1. **Lock the foundation** — first prompt creates the schema, the routes, the basic UI. Verify, commit.
2. **Step through additions** — one feature at a time on top. Each step is a small prompt against a known-good base.

This is the pattern most successful vly users converge on. It's the difference between "I tried for 20 minutes and gave up" and "I shipped a real feature in 45 minutes."

## Tips

<Tip>
  **Take breaks every 10 prompts.** Decision fatigue is real. Step away for 2 minutes; come back, you'll catch issues that were invisible mid-flow.
</Tip>

<Tip>
  **Name your revisions.** When you commit, give the revision a meaningful name. Future you (or your teammate) will thank you when scrolling [version history](/building/version-history).
</Tip>

<Tip>
  **For risky changes, branch first.** [Branching](/building/branching) gives you a clean undo even for changes that span multiple commits.
</Tip>

## Related

<CardGroup cols={3}>
  <Card title="Plan mode" icon="workflow" href="/building/plan-mode">
    The cheapest insurance against misunderstanding.
  </Card>

  <Card title="Version history" icon="history" href="/building/version-history">
    Roll back when needed.
  </Card>

  <Card title="Debugging prompts" icon="bug" href="/prompting/debugging-prompts">
    What to do when the agent gets it wrong.
  </Card>
</CardGroup>


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.