Digital Garden

How to use AI well

I was already using AI at work when I started cooking with my mom using Cooking with Claude. That was the beginning of using it to improve my life outside work. Since then, the two have kept feeding into each other: things I learn at work find their way home, and things I try at home change how I use AI at work.

Cooking gave me somewhere to start. We worked on menus, adjusted portions, and put together shopping lists. Recipes and preferences stayed in the project, so we could build on them the next time we cooked.

That grew into Rita, my personal assistant, who runs on OpenClaw. She has access to tools and files on my computer, and we talk through WhatsApp. Other people in my life can talk to her there too.

Someone cooking at home can ask her what to make, tell her an ingredient is missing, or ask for help with a step in a recipe. When there was no pesto for a pasta dish, she adapted it to garlic and lemon. When a recipe called for peeled peppers, she explained how to roast and steam them to loosen the skin. Those conversations happen directly with her.

She also helps check what actually arrived in a grocery delivery, clarify availability with shops, and pass along household instructions. People can ask follow-up questions and correct her as they go. I can give her a task that involves a conversation with someone else, then get the answer back.

Teaching it how you want things done

Across both work and home, I've had to get better at explaining what I want. Give AI the information it needs. Show it a good example. When something is wrong, explain why. When something works, say "always do it like that" and have it save the instruction for next time.

With Rita, those instructions live in files she reads. They include food preferences, how to handle recipe requests, and when to check with me. They change as we use her and discover what needs adjusting.

At Recurrente, the payments company where I build software, I have Claude Code and Codex review each other's work and send findings back for corrections. The same questions matter at home: did it understand the request, does the result work for the person using it, and what should it keep for next time?

A useful place to start is something you keep repeating. The third time you do it, try handing it over. Give yourself time to work through the first attempt and check the result. Keep the instructions that made it useful.

Bringing this to WhatsApp

At Recurrente, we're building Cliq, a WhatsApp assistant you can use without setting up something like Rita yourself. It will remember preferences you ask it to keep and help with paying bills, buying things at the pharmacy, and booking appointments. It will ask for your approval before moving money.

I also help teams put AI to use in their own recurring tasks. If there's something you're doing for the third, tenth, or hundredth time,

leave me a message.

© 2026 Max Holzheu