How to Actually Get Good at AI

I get asked this a lot: “How do I learn AI?” And the answer people expect is a course recommendation, a book list, or a YouTube playlist. I can give you some of those, but they’re not actually how I learned. And I don’t think they’re how you’ll learn either.

Here’s the truth: I learned AI mostly by using it. Constantly, stubbornly, sometimes badly. And I think that’s the only way that really works.

Why Most Learning Approaches Don’t Stick

Think about how people typically try to learn AI:

1. Watch a 4-hour YouTube tutorial about “prompt engineering”
2. Take a Udemy course with 12 modules
3. Read articles about “10 AI tools that will change your life”
4. Follow 50 AI influencers on social media
5. Feel overwhelmed. Do nothing.

The problem isn’t that these resources are bad. Some of them are genuinely excellent. The problem is that passive consumption doesn’t build the kind of understanding you need.

AI isn’t like learning a software tool where you memorize the menus and shortcuts. It’s more like learning to cook. You can watch cooking videos all day — you’ll learn some vocabulary, pick up a few tricks, maybe understand some theory. But you won’t know how to cook until you’ve stood in the kitchen, burned a few things, and developed an intuition for what works.

You need mental models, and mental models only form through experience.

What I Actually Did

Let me be honest about my own journey, because it was messy and I don’t want to pretend otherwise.

I took one really good course

I took an AI course at MIT. It gave me a solid theoretical foundation — what types of AI exist, how machine learning works, how these systems are actually built, what they can and can’t do. This kind of foundational knowledge is valuable because it doesn’t go stale as quickly as tool-specific knowledge does.

I’m not saying everyone needs an MIT course. But some theoretical grounding helps. It gives you a framework to hang everything else on. You understand why things work, not just that they work, and this makes you much better at adapting when tools change — which they do constantly.

I experimented relentlessly

This is where the real learning happened. I set myself challenges. Some were work-related: “Can I use AI to analyze our competitors faster?” Some were purely creative: “Can I make a podcast about my company with AI?” Some were just for fun: “Can I generate a video of my concept?”

Not everything worked. Some experiments were complete failures. But each one taught me something — about the tools, about how to communicate with AI, about what’s possible and what isn’t. And this knowledge compounds. Insight from experiment A connects with something from experiment B, and suddenly you see possibilities you couldn’t see before.

I optimized my daily work

The most consistent learning came from looking at my actual daily tasks and asking: “Could AI make this better?”

Email drafting. Meeting preparation. Document analysis. Presentation creation. Data interpretation. One by one, I pulled these into my AI workflow. Some tasks turned out to be perfect for AI. Others weren’t worth the effort. But trying each one taught me something about where AI fits and where it doesn’t.

This is also where the biggest personal payoff lives. Every task you successfully move into an AI-assisted workflow frees up time and energy for harder, more interesting work. It’s not about being lazy — it’s about being strategic with your attention.

I expanded what I thought I could do

This one surprised me. AI didn’t just make my existing work faster — it opened up entirely new things I couldn’t do before.

I’m not a designer, but I can create professional visuals now. I’m not a video editor, but I can produce video content. And despite not being a developer by trade, I’ve built tools and systems that work. Each new capability I picked up opened more doors, which led to more learning, which opened more doors.

That’s the compounding effect at work. Your AI skills don’t just add up — they multiply.

What I’d Skip

Since we’re being honest, here’s what I found to be a waste of time:

Most AI video content

Unpopular opinion: most AI YouTube videos have about 20% of the information density of a well-written thread on X (Twitter). A 15-minute video often contains 2 minutes of actual insight, padded with intros, outros, “smash that like button,” and repetitive explanations.

I’m not saying all video content is bad. Some creators are genuinely excellent. But if you’re the kind of person who prefers to learn quickly and efficiently, seek out written content — blog posts, threads, documentation. You can scan it, skip what you already know, and spend your time on the parts that actually matter.

Prompt libraries

I wrote about this in another guide, but it bears repeating: prompt libraries are like memorizing phrases in a language you don’t speak. They give you a false sense of competence without building real understanding.

Instead of memorizing “Act as a marketing expert with 10 years of experience in B2B SaaS…” — learn why giving AI a role helps. Learn why context improves output. Learn how to have a productive conversation with AI. That understanding transfers to every tool, every task, every situation. Memorized prompts only work for the exact scenario they were written for.

Trying to follow everything

The AI space moves absurdly fast. New tools launch daily. Models get updated weekly. If you try to keep up with everything, you’ll spend all your time reading and none of it doing.

Pick a few trusted sources. Follow people who actually build things with AI, not people who only talk about it. And spend more time experimenting than consuming. You’ll learn more in an hour of hands-on use than in a day of reading about what other people did.

What I’d Actually Recommend

Start with one tool, use it every day

Don’t try five tools at once. Pick ChatGPT or Claude, and commit to using it every single day for at least two weeks. For everything. Even for things where you think “this would be faster to do myself.” You’re building a habit and developing intuitions, and that requires repetition.

Invent challenges for yourself

Don’t wait for the perfect work task. Make up challenges:

• “Can I use AI to plan my next vacation in a way that’s actually better than doing it manually?”
• “Can I create a presentation about a topic I know nothing about?”
• “Can I build a simple tool that solves an annoyance in my daily life?”
• “Can I write a blog post in half the time using AI as a writing partner?”

The sillier or more creative the challenge, the more you learn. You’re stress-testing the tools in ways that reveal their strengths and limits. And you’re building that intuition for what AI is good at — which turns out to be useful for everything else.

Learn to have a conversation, not write a query

The biggest mindset shift for most people: stop treating AI like a search box. Start treating it like a conversation with a very smart, very fast colleague.

• Start vague, then get specific
• Push back when the output isn’t right — AI agrees with you too much by default
• Ask the AI to critique its own work
• Ask it to ask you questions before it starts working

That last one is powerful. Try saying: “Before you answer, ask me five questions that would help you give a better response.” You’ll be amazed at how much better the output gets when the AI understands what you actually need.

Build something tangible

At some point — ideally early — build something you can show someone else. A document. A presentation. A simple app. An analysis. Something real.

This forces you out of the theoretical and into the practical. It reveals gaps in your understanding. And it gives you something to be proud of, which fuels motivation to keep going.

Get a theoretical foundation (eventually)

Once you’ve been hands-on for a few weeks, consider getting some structured knowledge. Not a “prompt engineering bootcamp” — something that explains the fundamentals of how AI actually works. What’s a language model? What’s a token? What does “context window” mean? Why does AI sometimes make things up?

This knowledge gives you a deeper understanding of why certain approaches work. It makes you better at troubleshooting when things go wrong. And it helps you evaluate new tools and capabilities as they emerge — because you understand the underlying principles, not just the surface features.

The Compounding Secret

Here’s the thing nobody tells you about learning AI: the beginning is the hardest part. The first week feels clunky. The second week is slightly better. By the third week, you start getting genuinely useful results. By month two, you can’t imagine working without it.

And then it compounds. Every skill you build connects to skills you already have. You start seeing patterns. You develop intuitions. Eventually, you begin combining AI with your own domain expertise in ways that are unique to you — because nobody else has your specific combination of knowledge, experience, and perspective.

That combination is what makes AI powerful. Not the AI alone, and not your expertise alone — but the two together. The AI brings speed, breadth, and tireless execution. You bring judgment, context, and taste. Neither is complete without the other.

So start now. Pick a tool. Give it a challenge. See what happens. The returns start small and grow fast — but only if you actually begin.