How to practice AI?

I wanted to keep my impression short. We have been watching this AI tooling grow and we are amazed. And then we try to apply the latest version of something like OpenClaw, which uses all the models at the same time and every data extraction pipeline. I truly believe this is not the best way for a professional to stop and understand AI, when many people still can’t use a simple ChatGPT in a minimally interesting way.

That’s why I’m more in favor of using Cursor, Claude Code and Codex, these better, agentic, coding versions, so you build muscle on the foundation of the technology. Avoid using skills, avoid using complex plugins with thousands of tips from other people. And more: go back to the roots, try to solve a problem that you yourself have. That simple little system you always wanted to program, always wanted to solve, and didn’t have either the time or the will.

Remember when a new framework, a new programming language, came out? What did you do? You looked around and saw: hey, I think I can apply it here. And you went, studied, applied and learned. And guess what? Often it wasn’t that way. Often there were other, better steps to take, more ready to solve that mission. And that’s fine, it was a way for you to understand what’s under the hood. It’s the same analogy as someone who learns a modern SPA with React and understands nothing of vanilla HTML and JavaScript. This is the moment to use the pure technology.

For those using AI for something other than coding, my idea is that you don’t use it to create content. Don’t create. Use it for refinement, to find what you need, to execute what you already know needs to be done and where the data and content to be stitched together are. Creating a blog article from scratch, or a social media post from scratch, giving only a topic and a tone of voice, is to me a gross mistake. You need to be specific about what you want. That is: the content, the idea and the tone of voice, not only the tone of voice, it’s all yours and it’s all given. What it can help you with is polishing what was already in your mind, and not only in your mind, but within your reach.

This belief that AI will solve a creative problem in a simple prompt will produce that bland rice and beans we’ve been seeing around, the AI slop. In the best-case scenario, authentic content and intentional software will become increasingly scarce. Now is the time to put yourself into the result the AI is going to generate. For that, you need to be completely involved with the agents. This craze of wanting to create more agents, all autonomous, based on half a dozen paragraphs that are a specification, has a limit. You need to be in the loop, not necessarily to check that it’s not going off the rails, but because it needs your intention, information and genius. Without that, AI will only execute the basic, minimal tasks, or will generate the content everyone is already tired of seeing.