LLMs are amplifiers. They amplify your thinking. They amplify your prompts. They amplify your direction.

But here’s what nobody tells you: if there’s nothing to amplify, you get nothing useful back.

The Problem

You ask the AI a question. It gives you a fluent, confident answer.

But you have no idea if it’s right.

You can’t evaluate it. You can’t tell if it’s missing something crucial. You can’t catch the subtle errors mixed in with accurate information.

So you use it anyway. Or you ask another question. And another. You’re having conversations with AI, getting outputs, feeling productive.

But nothing sticks.

Tomorrow you’ll ask the same questions. Next week you’ll make the same mistakes. The information passes through you without becoming part of your thinking.

You’re not learning. You’re not building. You’re just… prompting.

Why This Happens

Most people approach AI the same way they approach learning: consume and hope.

Read enough articles, knowledge transfers. Watch enough videos, skills develop. Prompt enough questions, understanding emerges.

This is the cramming mindset - the belief that consumption equals learning, that exposure equals understanding. It’s decision avoidance disguised as productivity.

And AI makes it worse, not better.

(For a deeper dive on the cramming mindset and why it sabotages learning, read If You’re Not Careful, Your Notes Will Undo Your Learning.)

The short version: real understanding requires you to decide what you think. To write about it in your own words. To form mental models. To crystallize your thinking.

Without that foundation, you have nothing to bring to an AI conversation. And nothing to evaluate what comes back.

What Effective AI Usage Actually Requires

Here’s what separates people who get value from AI from people who just generate fluent emptiness:

They did the thinking work first.

This means they have:

1. Intent (To Know What They Want)

Prompting is directing. But directing requires knowing where you’re going.

If you haven’t thought deeply about a topic, you can’t prompt effectively about it. You don’t know what you’re asking for.

Vague understanding → Vague prompts → Vague outputs

“Explain machine learning” gets you Wikipedia.

“I’m thinking about ML as pattern matching rather than reasoning - where does this framing break down for my debugging system?” gets you something useful.

The difference is crystallized thinking. You have to show up with something.

2. A Foundation To Evaluate Output

LLMs hallucinate. They state wrong things confidently. They mix errors with accuracy.

How do you catch this?

You need your own understanding to compare against.

Without a foundation, the AI could tell you anything. Fluent output feels like correct output. You’ll confidently share misinformation because you had no filter.

With a foundation, you notice when something’s off. “Wait, that contradicts what I know…” or “That’s technically right but missing the point…”

Your thinking becomes the quality filter.

3. Context The AI Doesn’t Have

LLMs don’t know:

  • Your specific situation
  • Your projects and constraints
  • How this connects to your other work
  • What actually matters to you

This context comes from your crystallized thinking. When you know what you’re building and why, you can bring that to the conversation.

“How do I build a web app?” → Generic response

“I’m building a content pipeline on Cloudflare Workers that needs to handle LinkedIn rate limiting given my posting schedule. Here’s my current constraint…” → Response that addresses YOUR situation

The AI can’t read your mind. You have to bring the context.

The Two AI Traps

Without doing the thinking work first, people fall into two patterns:

Trap 1: Always Prompting, Never Processing

Question after question. Answer after answer. Never stopping to crystallize what any of it means.

The AI becomes a consumption firehose. Same pattern as reading articles without synthesizing - just faster and more fluent.

You feel productive. You’re not. The information passes through without becoming part of your thinking.

Trap 2: Outsourcing Decisions

“Just ask the AI” becomes your answer to every uncomfortable question.

What should I build? What do I think about this? How should I approach this? Is this good enough?

The AI gives answers. But they’re not YOUR answers. You’ve outsourced the decision-making that thinking requires.

Worse: the more you outsource decisions, the less capable you become of making them. Your judgment atrophies. You become dependent on AI not just for information, but for thinking itself.

The Fix: Learn First, Prompt Second

Before you can use AI effectively on a topic, you need to do the thinking work on that topic.

  1. Consume the material - Read, watch, listen. Take notes on what the source says.

  2. Crystallize your thinking - Later, separately, write about what YOU think. In your own words. Not summaries - your actual understanding.

  3. Make decisions - What does this mean? How does it connect? Do you agree? Why does it matter? These decisions ARE the thinking.

  4. Then prompt from that foundation - Now you have intent, context, and the ability to evaluate.

The sequence matters. You can’t skip to prompting and hope AI does the thinking for you.

What This Looks Like

Without the thinking work:

“Explain Kubernetes to me”

Gets a generic explanation. Can’t evaluate it. Doesn’t connect to anything. Forgets it. Asks again next week.

With the thinking work:

“I understand Kubernetes as an abstraction layer that treats infrastructure as cattle not pets. I’m trying to figure out how this changes my mental model for debugging - in traditional servers I SSH in and poke around, but that doesn’t make sense here. What’s the equivalent debugging workflow, and where does my ‘SSH and investigate’ instinct lead me wrong?”

Gets a response that engages with YOUR thinking, addresses YOUR specific confusion, and builds on YOUR mental model. Can evaluate whether it makes sense.

The second prompt works because you showed up with something. You had a mental model. You had specific confusion. The AI has something to work with.

The Pattern

  1. Learn first - Consume, then crystallize through writing
  2. Prompt second - Bring your thinking to the conversation
  3. Evaluate with your foundation - Your understanding is the filter
  4. AI extends, doesn’t replace - Amplification requires something to amplify

The Uncomfortable Truth

You cannot prompt your way out of not thinking.

AI makes thinking MORE important, not less. Because now you have a tool that produces infinite fluent output whether or not you understand any of it.

The temptation to consume without synthesizing has never been stronger. The opportunity to avoid decisions has never been more convenient.

Resist it.

Do the thinking work first. Build a foundation of crystallized understanding. Then use AI to extend that foundation.

The AI amplifies what you bring. Make sure you bring something worth amplifying.