Try Something. Fail. Figure Out Why.

A friend in tech asked how to get comfortable with AI. Her assumption was that she needed to understand the technical foundations first.

That assumption used to be true. When I first started exploring AI, technical depth was the only way in. I took courses, learned Python, built simple image classifiers. But the tools have changed. You no longer need to understand how the models work to use them productively.

What you do need is a willingness to experiment and the patience to figure out why your experiments fail.

With generative AI, I started small. Researching companies, planning menus, finding products. Low stakes, just building a feel for what it could do. It didn’t have to be perfect: when I knew what I wanted but not how to ask for it, I’d describe the outcome and let the AI draft the prompt.

Then I tried something more ambitious. I was analyzing a few hundred LinkedIn posts for a client, tagging each by topic, theme, and format. The AI handled the first few posts beautifully. When I ran the full set, it returned placeholders. No errors, no indication anything had gone wrong. When I asked what happened, it said the environment had reset.

So I broke the dataset into smaller batches. That worked, until the model started hallucinating.

That failure lit up the next stretch of the path. I knew about context windows, but I’d never had to actively manage one. Hitting that wall gave me a concrete reason to learn how.

That’s how it’s always worked for me, with Quark XPress, HTML, Python, Figma. Try something. Fail. Figure out why. Move forward.

AI is genuinely different, but the approach to learning it doesn’t have to be.

So start even if you feel like you don’t know what you’re doing. Then pay attention to what goes wrong.

Originally published on LinkedIn