Building something, even a small internal tool or a personal project nobody else uses, rewires how you think about the relationship between intent and outcome. You stop being a consumer of other people’s software decisions and start developing a model for how those decisions get made.
That model is what you bring to AI-assisted work. Without it, you describe what you want in terms of features. With it, you describe what you want in terms of problems. The second framing produces dramatically better outputs.
I have colleagues who are excellent designers and have never written a line of code. The ones who’ve built something, even a small automation, even an internal tool with Claude Code, ask different questions. They push back on agent outputs differently. They catch wrong turns earlier.
It’s not the code. It’s the model you build while writing it.
The names will change
Every few months there’s a new list of the skills that will matter in the coming decade. Prompt engineering, context engineering, loop engineering, evals, whatever the current word is for the thing that was obvious six months ago and will be table stakes six months from now.
I don’t think any of them is the skill. The skill is whatever lets you pick up the next project and, despite working in an unfamiliar environment and making mistakes, figure out why they happened, adjust, then do it again when the environment moves underneath you.
Building something with AI is practice for that. Not because the specific tool matters, but because the people who’ll navigate this moment best aren’t the ones who know the most about AI right now. They’re the ones who’ll have built the best working model of it by the time it changes again. And the only way to build that model is to build with it.
So build something small. Something you actually need. Something with real constraints and a real user, even if that user is just you.
The thing you build doesn’t have to matter. The mental model you build in the process does.
Also in August: Build to Learn AI · What Building Teaches That Reading Cannot · The Sandbox Is How You Learn
Continues in September with Taste, Thought, and Trust Don’t Come From the Model, on why a product that works can still make people hesitate.
Daniel Arevalo is a Product Architect and strategic product design leader. He helps product teams turn AI capability into trusted, adoptable workflows, with a background in industrial AI, IT/OT cybersecurity, and high-stakes enterprise software for critical infrastructure.