Watch your app come alive
In the next few minutes, you'll watch Emergent's AI agent turn your description into a real, working app - no coding required.
Choose how to start: Build or Plan
The composer has a Build / Plan switcher (a pill-shaped toggle; ⌥P also toggles it):
- Build (the default), the agent starts working as soon as you send your prompt. No waiting on plan approval; it gets going immediately.
- Plan: the agent first presents its proposed approach as a plan proposal card with two options: Request changes or Looks good, let's build it. Nothing is built until you approve. Use whichever option suits you.
For your first build, Plan mode is the safer start, you see the full picture and can refine it before the main build work begins.
Plan mode lets you review and adjust what the agent intends to build before it starts generating code. Note that planning itself also bills per token, a plan turn is much smaller than a build turn, but it isn't free (see How credits work).
Plan mode is not available in Brainstorm and E3 sessions.
Watch the agent work
Once the build starts, you'll see a live stream of steps in the chat:
- Planning - breaking your idea into components and features, the agent might ask you questions like "Do you need users to login?" or "Any additional info?" or questions about the features of the app here.
- Scaffolding - creating the project structure and installing tools
- Implementation - writing the code that makes your app work
- Testing - checking for errors automatically
The first build usually takes a few minutes. The agent works autonomously - you don't need to approve every step.
The platform automatically saves checkpoints of your app as the agent works. If a change breaks something, you can roll back from the chat timeline, note that rolling back erases the messages and code changes (optional) after the point you choose, so you roll back to a state, not "undo one thing." See Checkpoints: undo anything.
Your first preview
As soon as the agent finishes, a live preview appears. This is your app, running in real time. You can click through it, test features, and see exactly what you built.
When the build round completes, the agent will print a short summary in the chat recapping what it created or changed - then it invites you to open your app and try it out. The preview updates automatically whenever the agent makes changes - no refresh needed.
Try interacting with your app like a real user would: tap buttons, fill in forms, navigate between pages. If something doesn't look or work the way you expected, just describe the change in chat and the agent will fix it.
The preview is separate from your live app (production). Changes you see in the preview won't appear in production until you explicitly Publish (first time) or Re-publish (updates). The preview is your testing playground; publishing is when it goes live for real. See Put your app live.
If something looks wrong
If the agent stops or the preview doesn't match what you asked for, you have options:
- Describe the issue in chat and the agent will iterate. Be specific: "Move the search bar to the top" works better than "the design of the app looks off."
- Roll back if a recent change broke something that was working. Use the Rollback button in the chat timeline on the message you want to revert to.
- Ask the agent to explain what it built if you're not sure why something works the way it does.
The agent learns from your feedback and adjusts as you go.
Start small. Your first build should be something you can describe in a few sentences. You can always add more features later - it's faster to build in layers than to fix a big, complicated first attempt.
Go deeper: The chat-to-publish flow · Previewing & iterating
