Claude Code's creator says a good AI prompt needs just three things

Boris Cherny, the creator of Claude Code, says a good AI prompt comes down to three things: say what you want, say how much effort the AI should put in, and give it a way to check its own work. XDA writer Mahnoor Faisal looked at the prompt that set off the discussion and walked through how it fits those three tips.

What started it

On October 6, 2026, Cherny posted on X about an interactive website he built with Claude. It accompanied an episode of the Acquired podcast about Home Depot. People ended up talking less about the site and more about the prompt behind it. It was plain and chatty, like a message to a coworker. There were no special tricks or long templates.

The three tips

• Say clearly what you want and what the finished result should look like.
• Say how much effort the AI should spend on the task.
• Give the AI a way to verify its own work.

The first tip does not mean writing a long essay. It means the AI should not have to guess what you have in mind. The second relates to something Anthropic, the company behind Claude, has been pushing: letting users control how much time and computing effort a model spends on a task. A simple question does not need minutes of thinking. A hard one may benefit from more.

The third tip is one Cherny repeats in his Claude Code setup advice. If you ask the tool to build an app, ask it to run tests so it can tell whether the app works. Otherwise, you are trusting that it got things right.

What the prompt looked like

Cherny asked Claude to make an artifact, meaning a finished piece of content, that narrates the whole Home Depot episode. He wanted it beautiful, interactive and focused on the business lessons. He suggested watercolor illustrations, historical photos and a look inspired by The New Yorker or The New York Times. He also said to use lots of tokens, which are the units AI tools use to measure the work they do, and to keep iterating until Claude was proud of it.

The prompt is specific about the goal and the look and feel. The excerpt available here does not show how Cherny handled the verification step, and it cuts off before the results of the test.

What to take from it

Prompting advice often swings between give the AI everything and keep it short. Cherny's approach sits in the middle. Be clear about the goal, say how hard the AI should work, and ask it to check itself. That works in everyday chat tools too, not just for programmers.