Learn to Think With AI Before You Try to Build With It — Jake Van Clief's Best Advice for Beginners

I just watched Jake Van Clief's April Q&A live and had to break it down for you because it answered something I get asked constantly: where do I even start with AI?

Jake did a freestyle live where he worked through a Google Form's worth of beginner questions from his 13,000-person community. No script, no plan. Just him answering the questions he sees most often — and the through-line across almost every single one of them was the same.

Learn how to think with these tools before you try to build with them. The technical stuff fills itself in once you know what questions to ask.

The Question Everyone Gets Wrong

The most common beginner mistake is not a lack of coding ability. It is treating AI tools like vending machines — you put in a request, you expect an output, and when the output is off, you assume the tool is bad or you are doing it wrong.

Jake put it this way: before you touch syntax, before you open Claude Code or Codex or any AI agent, understand why those tools exist. What problems were they designed to solve? Why do folders exist? Why did someone design the system structure your computer runs on?

These are questions most people take for granted. But understanding the answers is what separates someone who can direct AI meaningfully from someone who just prompts and hopes.

You Have to Be Excited About the Problem

This was the part that really landed for me: Jake told someone asking about getting started in university to pick problems they are genuinely passionate about — because if you are not excited about what you are learning, you will not actually learn it.

I have seen this exact thing in the communities I have been a part of. I am the "AI person" in a lot of rooms. I have taught people how to use AI tools for their businesses, walked them through automation setups, helped them figure out what Claude or ChatGPT is actually good at versus where it falls flat. And the people who actually stick with it and get results? They started from genuine curiosity, not from "I should learn this."

The ones who approach it like homework — a box to check — stall out every time.

On Claude Code Specifically

Jake got a great question in the live about Claude Code and whether beginners need to understand what is happening in the back end. His take: you do not need to read every line it writes. But you need to understand what you are asking it to do.

The bottleneck is not the technical output. The bottleneck is the quality of your thinking. Claude Code, Codex, any agentic tool — they amplify whatever you bring to the conversation. If you bring clear thinking and a well-formed problem, the output is good. If you bring vague direction and a fuzzy goal, the output will be vague and fuzzy.

This is true whether you are a developer or someone building a simple automation. The tool is not the limit. Your ability to think through the problem is.

How to Actually Start

Based on Jake's Q&A and what I have seen work:

1. Start with why, not what. Before learning any specific tool, understand the category of problem you are trying to solve.
2. Pick something close to a problem you already have. Abstract learning is forgettable. Learning through a real problem you care about sticks.
3. Understand fundamentals before you automate. One module of basic computing concepts — what folders are, how code gets executed, what an API does — will make everything else click faster.
4. Use the tools to learn the tools. Jake's community uses AI to understand AI. Ask Claude to explain why something works. Use the assistant as a teacher, not just an executor.

The Bottom Line

Jake Van Clief gave a genuine, unfiltered answer session that I think every AI beginner should watch. The core message is simple: slow down, build understanding first, and let the technical pieces follow. That is the same advice I give in my own communities, and it works every time.