What Goes Into an Idea

A confession about where ideas actually come from

Most of the tools I've ever built — and most of the ones I trust — didn't start with a roadmap. They started with me noticing something I already do without thinking about it.

I would read a paper, underline a sentence, copy it into a notebook, and then forget which paper the sentence came from. I would have three browser tabs open with three different research questions, and I would lose the connection between them by lunch. I would finish a long article, close the tab, and have nothing left but a vague feeling that something in there mattered.

Nothing about that was a software problem. It was a system problem. The work was already happening — I was just doing it badly, and losing most of it as I went.

That gap — between the work you actually do and the system you do it in — is where every idea I care about starts.

Step one: stop, and pay attention to your own hands

The first step isn't 'what should this be'. The first step is 'what am I already doing, and why does it work when it works'.

If you watch yourself work for a week, you'll see patterns you couldn't have predicted. The notebook always opens on the left. The notes always get longer, not shorter. The pages you reach for are the ones near the middle of the pile, not the ones on top. The thing you come back to is rarely the thing you wrote down first.

None of that is in a feature list. All of it matters.

If you skip this step, you'll build something generic. And the reason it will feel generic — even if it's beautifully designed — is that it isn't actually for anyone. It's for the average user, which is nobody.

Step two: borrow the affordances of the physical world

Paper works for a reason. Notebooks work for a reason. Sticky notes work for a reason.

When you move a workflow into the digital realm, the temptation is to flatten it. To replace a stack of papers with a list. To replace a wall of index cards with a search box. To replace a desk you can spread things across with a single scrolling page.

That's almost always a mistake. The reason the physical version worked is exactly the thing you just threw away.

When I sat down to design NoteDex, I kept one question in front of me the whole time: 'What does this look like on a desk?'. Not 'what is the data model'. Not 'what does the navigation look like'. What does this look like on a desk.

The answer turned out to be: a lot of small cards, each with one idea on it, that you can spread out, rearrange, group, and pick back up. The whole product follows from that one observation.

Step three: let the agent meet you where you are

There is a moment in every workflow where the bottleneck is not your effort. It's your memory.

You wrote a note three weeks ago. You have a vague feeling it matters. You can't find it. You give up and write a new one.

AI doesn't fix this by being smarter than you. It fixes this by being patient in a way you can't be. It can read everything you've written. It can read everything you've read. It can sit in the background and wait for the moment when the new thing you're working on actually connects to the old thing you forgot about.

That moment — when a tool suddenly surfaces something you would have lost — is what the entire field is for. Not productivity. Not speed. Continuity.

If a tool can give you back one forgotten idea per week, it's done more than any productivity hack ever will.

Step four: build for a specific person, not a category

Every product I trust was built for someone the maker knew. A partner. A friend. A coworker with a specific job.

When I built the first version of NoteDex, I was building it for myself and one other person. Every decision was checked against 'does this help us' — not 'does this help everyone'.

The 'everyone' version of a tool is always worse than the 'someone specific' version of a tool. The 'someone specific' version has opinions. It makes choices. It does some things on purpose that other people won't like. That's a feature.

If your tool could be equally loved by any user, it's not a tool. It's a survey.

What doesn't go into an idea

Roadmaps don't go into an idea.

Market research doesn't go into an idea.

What your competitor launched last quarter doesn't go into an idea.

The next round of funding doesn't go into an idea.

None of those are bad things to know. They're just not the source. The source is what you notice when you stop trying to be a founder for a minute and just pay attention to your own hands.

How to actually start

If you want to build something real this week, here's the exercise.

Pick one thing you do every day that you can't quite explain why you do it that way. The way you organize your photos. The way you read papers. The way you decide what to cook for dinner.

Watch yourself do it for three days. Write down what you actually do, not what you would say if someone asked. Notice the parts that are awkward. Notice the parts that work despite being awkward. Notice the part where you almost always lose the thread.

That last part — where you lose the thread — is your product. Build for that exact moment. Name it. Make it the center of the design. Everything else flows from there.

The best ideas aren't breakthroughs. They're observations you've been carrying around for years that finally got a place to land.