Loading
Prepare for Email to Become the Default Login ExperienceRead More
Builder Central (Beta) Best Practices

Builder Central (Beta) Best Practices

Get better results from Builder Central: describe what you want clearly, keep your data private, and refine effectively when a result isn't what you wanted.

Describe What You Want to Build

The clearer your description, the closer the first build lands to what you want. Keep these tips in mind as you write your prompts.

  • Be specific about the goal. Say who uses the app and what they need to do, not just the app type.
  • Start small, then build up. Describe the core app first, then add features by refining in plain language.
  • Review the project details whenever the Build Agent shares them. Confirm the objects and behavior before you tell it to go ahead and start building.
  • Treat the result as a first draft. Expect to refine it, and check everything before you deploy.
  • Be specific about what to change. Name the exact element and change you want, or select it in Design, so a new request doesn't undo earlier work.
  • Point instead of describing, for small visual tweaks. Select an element in Design and adjust it in the property panel, rather than describing where it is.
  • Show the difference specificity makes. Instead of "Build a returns app," describe the outcome: "Add a topic to my agent that handles product returns. The agent should ask for the order number, check whether it's within 30 days, and, if so, start a return flow."

Keep Your Data Private

Builder Central is designed to keep your data protected. Keep these points in mind.

Your personally identifiable information (PII) is protected automatically. Builder Central masks PII before it's sent to the AI model, and this protection is on by default, so there's nothing to turn on. The masking applies to the prompts the Build Agent sends while you build. Salesforce doesn't apply this masking to the runtime agents you build.

Your data stays yours. Salesforce doesn't use your data to train its large language models (LLMs).

Refine Your Results

If a result isn't what you wanted, a few habits help you get back on track.

  • Change one thing at a time, so you can see what each request does.
  • Describe the outcome you want, not the problem. Say what the app should do, rather than what went wrong.
  • Be specific when a result is close but off. Name the element and the change you want.
  • Start a fresh description if the chat stops making progress.
  • Let the current step finish before you make another request. If the Build Agent returns an error message, paste the message in the chat and ask the Build Agent to fix it. Pasting an actual error message is different from describing what's wrong with a result, where it's better to state the outcome you want.

What to Test

Put Builder Central through the work you'd really do, and note what works well and where you'd want it to do more.

  • Test a real scenario. Build an app or agent you'd actually use, not just a sample, so you see how Builder Central handles your work.
  • Check the project details. Confirm that the details the Build Agent proposes match what you asked for, and note where they miss.
  • Preview the result. Confirm that the app looks and behaves the way you expected.
  • Refine in plain language. Confirm that you can make the changes you need just by describing them.
  • Review agent responses. Watch for a response that's too technical, or a refusal to do something the agent should handle. If you see this, refine your prompt to guide the agent, and note it so you can share feedback.
 
Loading
Salesforce Help | Article