June 28, 2026 · 7 min read
A Path to Think: Onboarding Into a Spatial Thinking Environment
Every new tool has a first enemy: the blank screen. Now make the screen spatial — the user isn't filling a document, they're placing nodes in space. Now make it radial, where everything orbits a center and position itself carries meaning. Each layer multiplies the problem, and there's no established playbook for any of it. There's no canonical pattern for onboarding someone into a spatial, radial thinking system the way there is for a signup form or a settings page. You don't look it up. You build the intuition by iterating, and this is how far that's gotten me.
What I've found along the way is that the issues are the asset. A hard onboarding problem rarely resolves into a patch — it resolves into a better architecture, a sharper experience, a new feature, or, best of all, a feature that sidesteps the original problem entirely. This is the story of being wrong twice, and what each wrong turn handed me.
Being wrong twice
First, I shipped a /try version of the map and let people create nodes with no help at all — here's a blank canvas, fill in your own categories, go. The result was flat. People didn't want to sign in, and most didn't even seem to understand what had just happened in front of them. A blank map doesn't explain itself; people pattern-match it to a mind-map and leave.
So I added AI as the way in — you say what's on your mind, and the map starts coming to life around it. That did noticeably better. More people stayed, and more of them stayed longer; the best single session was about ten minutes, which for a cold stranger off a link is a small miracle.
Then I posted it to Reddit, and about ten comments came back saying the same thing: we want templates.
My first reaction was a little defensive — the AI already solves the template problem. Tell it what you're working on and it builds the structure for you; what's a template next to that? But the comments kept nagging at me, and eventually I flipped it around: maybe I'm the one thinking about this wrong.
Here's what I landed on. I don't want to build canned project templates — pick "marketing plan," get a marketing plan. That's the part I keep refusing to do, because it's the copy-a-finished-thing move and it's the opposite of the point. But the people asking weren't really asking for my templates. They were asking for a way in. A starting shape. Something that makes the blank stop being blank.
They were asking for a path to think.
A path in order to think
That's where thinkpaths came from — literally, a path in order to think.
There are two pieces. The first is the starting structure: for the system to work, a project gets laid out as twelve radial categories around a center — twelve places to think from. You can build those categories yourself if you've already got the idea and its shape in your head. Or you can just say whatever's on your mind right now, and the AI figures out the rest and generates the twelve categories for you. Not a finished project. Not answers. Twelve doors, all of them empty, each one a place you can start thinking from.
The second piece is the thinkpath itself: off any one of those nodes, you fan out a way of thinking — a Socratic chain, a 5-Whys, a "steelman it, then find where it breaks." Here's one in full, straight from the registry:
// src/components/bubblemap/thinkplates.ts
//
// A thinkplate is a sequence of thinking PROMPTS, rendered as
// labeled ghost bubbles off a selected node.
// It is NOT content, and NOT a storage schema.
type ThinkplateBubble = {
label: string; // names the move
prompt: string; // the question that provokes it
};
const SOCRATIC: Thinkplate = {
name: 'Question it',
topology: 'chain', // each bubble deepens the last
bubbles: [
{ label: 'The claim', prompt: 'What are you actually asserting?' },
{ label: 'Assumptions', prompt: 'What has to be true for that to hold?' },
{ label: 'Evidence', prompt: 'What would convince a skeptic?' },
{ label: 'The opposing view', prompt: 'Where does the strongest counter live?' },
],
};
Look at the type. There's no answer field — anywhere in the structure. The bubbles arrive empty. The bubble is the question; you are the answer. That single absence is the whole philosophy, so let me slow down on it.
Not capture. Not an AI that thinks for you.
There are two obvious categories a tool like this could fall into, and both are wrong.
The first is capture — notes apps, transcription, mind-maps. You have a thought; the tool records it. The artifact is a history of thinking that already happened.
The second is the 2026 default: an AI that thinks for you. Prompt it, it produces the thinking. Clean demo — but it asks a stranger to trust an AI with their reasoning before they've seen a single payoff, and it quietly removes the one thing they came to do, which is think.
What I'm building is a third thing, and the empty bubble is how you can tell. The system hands you a structure to push against, but it can't put a word in your mouth, because what it gives you is a hole you fill. It doesn't record your thinking after the fact, and it doesn't do the thinking for you. It scaffolds the thinking while you do it.
Why "scaffolding" is the right word
I'd been reaching for verbs like "shapes your thinking," and they all felt wrong — they smuggle in coercion, like the software is driving and you're along for the ride. That's the opposite of the feeling I want.
Scaffolding fixes it, and not loosely. It's a real idea from learning theory — Vygotsky — for the temporary support that lets you do something you can't yet do alone, and is then removed. The freedom is built into the meaning:
- Scaffolding is not the building. Your thinking is the building.
- You can climb it or walk around it.
- It exists to be outgrown.
And the freedom is structural, not just tone — it's enforced in the architecture, not promised in the marketing.
The recommended starting point isn't a wizard or a forced first step. The comment in the component spells out the intent before any code runs:
// Recommended start — soft amber GLASS, not a solid primary button.
// Reads as an optional "here's a good place to begin," not a command.
// "recommended" + the soft fill keep freedom of choice intact.
Clicking a node never cascades edits — one click, one focused action, and that's it. And the AI's whole job is to build scaffolding, never the building. When it suggests a path, look at what it's even allowed to return:
// POST /api/projects/[id]/thinkpath-suggest
// Reads your node + its neighborhood, picks the shape (fan or chain),
// and writes prompts tuned to THIS thought.
async function suggestThinkpath(node: Node): Promise<ThinkplateBubble[]> {
// → [{ label, prompt }, { label, prompt }, ...]
// never an `answer`. the model gets you to the edge of the
// thought — and hands you the pen.
}
Every category and every prompt it makes lands empty: you fill it, you decide if it lives, you can ignore all of it and start somewhere else. The system holds the structure. You always hold the thought.
Never blank, never bossy
Concretely, here's what the environment does so you're never staring at a blank page — and never being bossed around either:
Empty prompts instead of a blank canvas. A blank page asks you to generate from nothing, which is the most expensive cognitive state there is. A thinkpath gives you something to react to, and reacting is cheaper and sharper than generating — but the prompt comes empty, so it's still entirely your thought.
Your past thinking stays in view. While you work on the next idea, everything you've already thought is right there on the map. You're not thinking alone in a text box; you're thinking next to your own accumulated reasoning.
The neighborhood appears on a glance. Hover any idea and its connections light up — lateral, associative recall, surfaced the instant you look, without leaving the thought you're in.
A good place to begin, never a command. The single hardest moment in any open environment is the first move, because every direction feels equally arbitrary. A soft, optional "start here" removes the paralysis without removing the choice.
Think more, not less
Here's what all of it is actually for.
There's a whole genre of tool right now built around the idea of a "second brain" — you copy information from somewhere, paste it in, store it, and let the AI fetch it back later. And every time I look at one I think: that's a great memory for your AI. But what about you? What about your thinking process in all of this? You didn't think anything. You moved a thing from one box into another box, and now it's stored so the machine can get it.
I don't want to build a place where you store thoughts you already had somewhere else. I want to build a place where you actually think them — and because you thought them, you remember them. It's not a copy-paste tool. It's not paste-a-URL-and-call-it-knowledge. The map is there to amplify your thinking, not to do it for you. Every decision in the system is biased toward getting you to think more, not toward letting you think less and go brainless because "the second brain has it now."
That's the whole bet, in one line: never blank, so you always have somewhere to start — never bossy, so the thinking stays yours.
Where this is going
My guess is that, taken far enough, "onboarding" stops being a separate thing at all. A mature environment doesn't have an onboarding mode — it's always quietly offering you a next move, whether it's your first hour or your thousandth. The generated categories, the recommended start, the thinkpaths: those aren't a tutorial you graduate from. They're early versions of an environment that always has scaffolding ready and never forces you onto it.
I haven't proven the conversion side yet — I've moved the front door, not won the argument. But the reframe is the part I'm sure of: the job was never to capture people's thinking, and never to do it for them. It's to build a place that's never blank and never bossy — and then trust people to think, because they were always going to be better at it than the machine.