I had a Notion database with hundreds of prompts. Organized by category. Tagged by use case. Some of them were three paragraphs long — carefully crafted instructions that took me 30 minutes to write.

Then came the real time sink: another hour going back and forth with the AI. Tweaking, redirecting, rephrasing. At the end of that hour, I still didn’t like what I got. So I’d print it out, rewrite it by hand, and wonder if I would have been faster doing the whole thing myself.

If this was how AI was supposed to multiply me, I was losing the game. And I needed that multiplier — I had just joined Landbot, a new company, a new domain, everything to prove.

The problem wasn’t prompting

The problem was treating prompts as the asset. They weren’t. My judgment was.

In January of this year, I didn’t know much about AI. Hobby-level at best. I had used ChatGPT like everyone else — asking questions, generating drafts, summarizing documents. Nothing that felt like a real shift in how I worked or thought.

I wasn’t starting from zero, though. I had been running a second brain in Notion for three years — capturing notes, organizing ideas, distilling what I learned from books and work. No AI involved. It was slow and tedious, but the judgment was already there: customer pains I’d seen, product principles I trusted, trade-offs I’d learned to make.

What I didn’t have was a way to make that judgment usable by the AI at speed.

Five months later

I’m proposing core architectural changes at Landbot — a conversational AI company. Not tweaking copy or automating reports. Proposing changes to how the product works at a fundamental level.

What changed wasn’t the model. It wasn’t a better prompting technique. It was what I brought to the conversation.

I stopped saving prompts and started saving judgment.

What that means in practice

I rebuilt my second brain into something the AI could actually work with. Not a folder of documents. A system where each book I read produces atomic notes with explicit trade-offs and confidence levels. Each customer interview becomes a searchable case. Each decision gets captured with the reasoning behind it — not just the outcome.

Today, the system has almost 400 permanent notes and 1,400 cross-references. But the numbers aren’t the point. The point is that each note carries a bit of judgment: what I trust, what I doubt, and where it connects.

When I sit down with Claude, I don’t brief it from scratch. I point it at a network of judgment that took three years to build.

Notes aren’t storage

They’re judgment made explicit — knowing which customer pain matters, which constraint is real, which UX principle applies, which trade-off is worth making, and which idea should be killed.

When an AI is generic, it produces generic output. But when you feed it curated judgment, something shifts. The output stops sounding like “an AI wrote this” and starts sounding like something you would have written. Except faster. And sometimes better, because it can work across more of your context than you can hold in your head at once.

The four-hour presentation

Two weeks ago I built a presentation with two working prototypes and a technical document referencing key papers and expert knowledge. In four hours. Alone. No design team, no engineering support. Just me and Claude, working from my knowledge system.

The prototypes followed UX principles I’d internalized from books I’d read months ago. They used Landbot’s design system and UIKit, reflecting technical constraints I understood from conversations with engineers. They solved a business problem I’d identified through real customer interviews and internal pain points.

I can’t prove the system caused this — it could be domain fit, timing, or experience. But I can trace exactly which notes fed which decisions.

The system isn’t magic. I still caught mistakes. I still had to decide what was good. But that’s the point: the AI didn’t replace my judgment. It made my judgment easier to apply.

The reaction

The important part wasn’t that people were impressed. It was that different teams recognized the same pain from their own angle.

The head of sales said the problem was crystal clear — a pain they themselves lived with every day. The tech lead: “Brutal.” The CEO liked the idea and asked how he could help bring it to life.

It’s not just me

A friend started using a version of my system two weeks ago. His take: “I don’t fully understand what you’ve built yet, but it’s clearly helping me learn branding. I feel more confident.”

He didn’t learn a prompt. He started building a judgment layer — and the AI finally had something real to work with.

Start here

You don’t need three years of notes to start. You need the discipline to make explicit what you already know. Most people have more judgment than they think. They just never wrote it down in a form the AI could use.

The prompt is a tiny step. The judgment is the road.

If you took away all your prompts and templates tomorrow, what would you actually have to give the AI? If the answer is “not much” — the fix isn’t another prompt library. It’s to start building the thing prompts were supposed to access in the first place: your judgment.