Cook with AI Get lifetime access
Best Practices

Build better products with AI by following proven practices.

Practical application-building and product-development best practices for planning features, shaping data, setting boundaries, shipping safely, and turning prototypes into real software.

How to use this section

Pick a best practice before you ask AI to code.

Each best practice focuses on one decision you make before or during a build. The goal is to keep the product clear enough that an AI agent can help without quietly inventing the wrong app.

Quick Tip: don’t rely on a model training data

I was reading this comment on Hacker News:

Automating Git commit messages with AI

One of the most useful applications of AI that has done a positive change in my life is automating Git commit messages.

Save your prompts as docs

When you finish a project, save the prompts you used.

Prompt in your native language

This one is for those non native English speakers.

Let the AI test in Cursor's browser

Cursor has a browser built in. The AI can open it, navigate to your local dev server, click around, take screenshots, and report back.

Paste the error verbatim

When something breaks, don't paraphrase. Don't write "the form isn't working." Don't say "there's some kind of database error."

Ask the AI to review its own work

When you're happy with what you've built, ask the AI one more question:

Use screenshots. Words are slow, pictures are fast

When something looks wrong, I don't describe it in a paragraph. I take a screenshot, drop it into the chat, and write five words.

Use the app, the next prompt will show up in your mind

The fastest way to find the next prompt is to use the app yourself.

Plan in plan mode, then build iterating

For anything bigger than a one line tweak, I usually split building into two phases:

Pick a simple stack

After 10 weeks of building with Cursor, I have a clear pattern. The projects that flew used a stack the AI could easily master. The projects that dragged use...

Write the first prompt like you're explaining the app to a friend at dinner

When you start a new project with AI, your first prompt sets the shape of everything that follows.

Deploy on day one, not at the end

One thing I see people make with AI built apps: they build the whole thing locally, then try to deploy at the end.

Build a CLI alongside the UI

Some projects I built with AI got a small CLI tool, in parallel with the web UI.

Use MCPs as backend superpowers

MCPs aren't just for design. They turn the AI into someone who can operate your whole stack.

Pull design from a real design tool via MCP

There's a step beyond "screenshot the reference" to create a design: you can hand the AI a real design from a real design tool.

Use AGENTS.md as the project's memory

Your conversation with the AI ends when you close the chat. The next session starts cold.

Member access

Best practices are part of the Cook with AI library.

Get product best practices alongside guides, tutorials, developer tools, and workflow material.

Get lifetime access