Field notes on teaching designers to use AI tools

---

Over the last few months I have been helping designers at Nubank get into AI tooling, mostly Cursor and the agentic coding tools that have been arriving so quickly. These are some of the things I picked up along the way. So lets get started.

Live with excitement first, complexity later

The first thing I noticed, and this surprised me a bit, is that the moment people understood what these tools could do was not when I explained how they worked. It was when they saw a final output, something running on screen that they made themselves.

Whenever I had the chance to show a designer a working prototype, or sit next to them while they shipped their first thing, you could see it click. That little moment of "wait, I just did that" is when most of them actually started to engage. The questions came after, not before.

So for me, the lesson is that excitement does most of the teaching for you, if you give it room. If you are the one helping someone get into this, my advice would be to chase that first smile. Pick something small, something they would actually want to use or show a friend, and get them there as fast as possible.

You do not need to understand it to use it, at least at first

A lot of designers I worked with kept getting stuck on the same things. They wanted to understand what React was, what GitHub did, whether they should use Vercel or Replit, what the difference was between a native app and a web app, what an API actually is.

Those are good questions, and I would say they matter eventually, but when you are starting they really do not. You can build something cool in Cursor without knowing any of that, and the curiosity to dig into the underlying concepts tends to show up naturally once you have something working in front of you.

So when someone came to me asking about frameworks before they had built a single thing, I would gently push back. Build something first. It does not need to be the right framework or scalable code or production ready, it just needs to exist and put a smile on your face. The minutiae becomes interesting once you have skin in the game.

Lead with questions, not with your way of thinking

This one took me longer to learn. When you know how to do something and you are teaching someone, there is this instinctive pull to just hand them your method, here is how I think about it, here is what I would do, here is the prompt I would write. It feels efficient in the moment, but for me it has not been the most useful way to help.

What I started doing instead was leading with questions. What are you trying to achieve, what did you try already, what did you expect to happen, why did you think that would work. The point was not to quiz them, it was to help them formulate the questions in their own head, so the next time they hit a wall they could formulate those questions on their own.

Because with agentic tools, the answers are mostly already there. The hard part is knowing how to ask, and if you do not know what to ask or what to search for, that is when you get stuck. So for me, teaching someone how to ask the right questions is way further along the path of adoption than teaching them any specific solution.

Curiosity is the only real prerequisite

If I had to pick one thing that predicted whether a designer would get good at these tools, it would not really be technical background or coding experience or seniority, it would be curiosity. The designers who got the furthest were the ones who would just try things, open Cursor, type something, see what happens, break it, fix it, try again. They did not wait until they understood, they just poked at it.

I think this is worth saying out loud because there is a lot of pressure right now, especially for people who feel behind on AI, to study before they play. Take the course, read the docs, watch the YouTube tutorial, and then never actually open the tool. For me, the people figuring this out fastest are the ones who give themselves permission to waste effort, build something useless, try a prompt that probably will not work, make a thing nobody asked for. We are all trying to make sense of all of this together, and that is much easier to do with hands on the keyboard than with the docs open.

A balance worth holding on to

Now, there is a counterweight I want to add, because I think it matters as much as the rest.

There is a real risk with these tools, which is that you start handing the rationale over to the model. Not the boring parts, the actual thinking, the problem framing, the decision about what to build, the judgment call about whether the output is any good.

The way I use these tools, and the way I encourage others to use them, is to keep the thinking and let the model take care of the boring parts. I am still doing a lot of the problem solving and a lot of the deciding. The model is helping me move faster, but it is not telling me where to go. When I see someone hand the rationale over to the LLM, when the prompt becomes "tell me what to build", that is when I get a bit worried, because the muscle of judgment is the part of the work that is actually yours.

Some final thoughts

If I had to summarize what I have been seeing, it would be something close to this. Help people feel the excitement before you explain anything. Let them build before they understand. Teach them to ask rather than handing them your answers. And whatever you do, do not let them hand the thinking over to the tool, because that is the part that matters.

Hopefully some of this is useful if you are also helping people adopt these tools, or if you are starting yourself. None of it is unique to AI, it is probably how most things are best learned, by being curious enough to play and disciplined enough to keep thinking while you do. The difference now is that the gap between trying something and shipping it is really small, which means there has never been a better time to be curious. Use that, help others use it too.