A workspace where people learn by doing, and AI guides instead of taking over

I set out to help people collaborate with AI on hard, ambitious tasks. The first version was an execution tool. It broke work into steps and let people hand each step to a person or to the AI so they could finish faster.

After watching more than fifty people use it, the data told me something I did not want to hear. When the goal is to finish, people will always choose automation over doing the work themselves. So I followed where users were actually pulling the product, away from finishing faster and toward learning how to do something. haX became a workspace for learning by doing.

Most AI today works like an answer machine. You type a prompt and it produces a result. That is useful when you only want the output, but it makes a poor teacher.

The people I care about are ambitious learners and improvers who want to challenge themselves. They do not just want the answer. They want to own the process and grow from it.

I treated AI collaboration as an execution problem. haX would break an ambitious goal into a clear workflow, then split the work between the person and the AI so the result stayed human where it mattered.

Across the closed beta, the same wall appeared again and again. The breakdown itself was valuable, and people loved seeing a fuzzy goal turned into a clear roadmap. But the step by step workflow slowed them down. When a step lacked guidance they stalled, and when it had guidance they leaned on the AI to decide and generate for them.

The beta data made that failure concrete. Most workflows never reached the end, and the automation people set up was rarely the thing that carried them there.

What people actually brought told the other half of the story. Their goals were framed around mastering and understanding a subject, not offloading a deliverable. Words like master, systematically, and crash course came up again and again, and in onboarding most people tagged themselves with academic and learning goals. Even the deliverable shaped goals were heavyweight knowledge work rather than tasks to automate.

So I stopped treating haX as an execution tool and rebuilt it around learning.

In the product itself, the screens barely changed. What changed was the logic underneath, the way a goal gets broken down. The learn-by-doing breakdown introduces new kinds of steps, Guide, Decision, and Review, that turn a flat task list into a path you actually learn from.

haX is now a workspace for learning by doing, built for people who want to achieve something more than a generated answer. Three moves carry the whole product.

I started haX as a designer working alongside an engineer. When that changed, I decided to keep going on my own and turn my Figma files into a real, working product using Claude Code.

That meant owning every layer of the product, from the interaction design to the front end to the AI logic behind it. This is the project where I stopped being only a designer and started thinking like a builder and a founder.

Here is haX from start to finish, from a goal to a guided path to something you genuinely understand how to do.

The biggest lesson was not a feature. It was learning to let the users, and my own hands, tell me what the product really was.