Prompting effectively

The clearer your request, the better the result. These habits get you further with less back-and-forth, and they stretch your token allowance because fewer rounds are wasted on rework.

Describe the outcome, not the implementation

Say what you want the app to do, not how to build it. "Show the newest posts first on the home page" works better than instructions about specific code. The agent decides how to get there, and it usually picks a sound approach when you describe the goal clearly.

Give context

Say who the app is for and what it should do. "A booking form for a small dental clinic, patients pick a date and time and leave their phone number" tells the agent far more than "add a form." Context about the audience and the purpose shapes better defaults for layout, wording, and behavior.

Make one change at a time

Ask for one focused change per request. Each result is then easy to check in the live preview, and you keep tight control over how the app grows. A single sweeping instruction is harder to review and harder to fix if part of it is off.

Start broad on the very first prompt to set the shape of the app, then switch to small, focused requests to refine it.

When something is wrong, describe exactly what is wrong

Instead of asking for a full redo, point at the problem. Include what you did, what you expected, and what actually happened.

• Vague: "it's broken."
• Clear: "The save button does nothing after I edit a row and click save. I expected the row to update in the table."

Specific descriptions let the agent find and fix the cause faster. See Troubleshooting.

Say what should not change

When you refine, name what should stay the same as well as what should change: "Keep the current layout, just change the button color to match the header." This keeps a focused request from touching more than you intended.

Example: from vague to clear

Vague:   Make a store.

Clear:   Build a store page that lists products from the connected
         database in a grid of cards, each with an image, name, price,
         and an "Add to cart" button. Add a cart drawer that opens from
         the top-right and shows the items and a running total.

Work with the way the agents build

The agents build in a loop. They reason about your request, make the change, then check their own work before calling anything done, and they keep going if more work remains. You get better results when your prompt sets that loop up to succeed.

• Set the shape first, then iterate. Use a broad first prompt to establish the app, then switch to small, focused requests. Each pass through the loop is then easy to review in the live preview before you steer the next one.
• For a big build, describe the stages. On larger, multi-part work you can pick Ameca, the project lead. It breaks the request into a plan and hands each piece to a focused specialist: A1 for the front-end, A2 for the back-end, and A3 for security and deployment. It then combines the results into one coherent change, so describing the stages and what matters most helps it plan.
• Let the checks work for you. Because the agents confirm their own changes, a completed step is usually one you can try right away. Review it in the preview, then describe the next change specifically.

See PlutoAI Agents to learn how the build loop and the specialists work.

Next steps