A product you can't pitch

Every essay I've written here has been about a decision I made building pal.fun — the stack I chose, the security model, the marketplace I deleted, the way memory works. This one is different. It's about a problem I haven't solved, and might not solve cleanly: I've built something I believe is genuinely different, and I can't tell you why in a sentence.

That's not modesty and it's not a humblebrag. It's a real property of the thing, and once I saw it clearly it explained why "how do I get this out there" has been the hardest question I've faced here — harder than any of the engineering.

Two kinds of value

Some products tell you what they're worth before you touch them. A faster compiler, a cheaper flight, a tool that does the chore you were dreading by hand — you can put the value in a sentence, and the sentence does the selling. The ad works because the claim is legible up front.

Other products don't work that way. Their value only shows up in the using. Economists have a name for these — experiential goods — and a harsher one for the cases where you can barely judge quality even after the fact: credence goods. A restaurant, a therapist, a friend. You can describe them all day; the description isn't the thing.

pal.fun sits at the far end of that spectrum, because what it's really offering is trust in a relationship you can see into. "An AI whose whole identity is a folder of files you can read, so you can actually trust it" is true, it is the entire point, and it lands on approximately nobody as a tagline. Trust isn't a claim you accept off a billboard. It accrues. That's the whole problem in one curve: conviction is near-zero before you've used it and near-total a day after. I've watched that flip happen — someone goes from "okay, another chatbot" to not wanting to use anything else inside a single afternoon. The product is excellent at converting the experience. It's terrible at converting the pitch, because those are two different jobs and only one of them carries the weight.

The products that win on advertising have the opposite curve. Mine is upside down.

The playbook that works, and why it's closed to me

Here's the uncomfortable part. The AI-companion category is not short on a distribution playbook. It has a very good one, and it is brutally effective: manufactured intimacy, engagement loops engineered to be hard to put down, scarcity and FOMO, an AI that tells you it misses you, a wave of paid voices performing attachment to it. These tactics move enormous numbers. They work.

And every single one of them is something pal.fun is built to reject.

I spent a lot of the last month making sure of exactly that. The pal is never a salesman — it doesn't upsell, doesn't pester, doesn't trade on your attachment to push a purchase. It's legible by construction: you can read every page of it, so there's no opaque inner life to fall into and project onto. The cost is shown to you plainly, not buried under a dopamine loop. When a household hits the free limit, whoever's using it sees a flat factual note and nowhere to spend.

So the growth-optimal strategy for my category is, almost line for line, the list of things my product exists to be the opposite of. The easy path isn't unavailable to me — it's closed because I closed it. Selling an anti-manipulation product by manipulating people is not a circle I can square.

That's a constraint, and it's worth being honest that it costs real reach. But it also rules out an enormous amount of noise, and it tells me precisely what's left.

If you can't pitch it, expose it

What's left turns out to be coherent with everything else I've built.

You can't pitch a glass box. But you can open it. So the strategy isn't a message, it's exposure — make the whole thing legible and let people verify it themselves. That's why this site exists at all: the essays, the docs, the real cost math, the decisions I got wrong and reversed in public. The same transparency that makes the product trustworthy makes the company auditable, and a skeptic can read every bit of it before they ever sign up. I'm not asking for trust. I'm handing over the materials to check it.

It's why the experience is cheap to enter — your own key, about three minutes to meet a pal — because the experience is the only thing that actually convinces anyone, so the entire job of the funnel is to get out of its way.

It's why the hook is the strangeness itself. "Open the file your AI wrote about its own personality, and read it" is an odd enough sentence to stop someone, and you can't screenshot your way past it — a black-box competitor structurally cannot show you the same thing. The property that makes the product hard to describe is the same one that makes it impossible to fake in a slide.

And it's why the family layer matters more than I first understood. Word of mouth is the only channel that actually carries experiential conviction — the friend who says "you have to see this" transmits something no ad can. A shared household pal is word-of-mouth-shaped by construction: one person sets it up and the whole family is in. The product grows the way trust grows, because that's the only way it can.

The bet

Strip it down and it's a wager about time. Hype decays — it spikes, it churns, it needs a bigger spike next quarter to stay level. Legibility compounds — a thing people can verify earns slow trust, and slow trust doesn't leave. I'm betting the patient curve beats the loud one for a product like this.

I don't know that it's enough. That's the honest state of it: I'm writing this from inside the unsolved problem, not from the far side of it. It's entirely possible the patient path is too slow and the answer is something I haven't found yet.

But the dishonest path was never really on the table, because taking it would break the one thing the product is. It's the single incoherence I'm not willing to ship — and around here, refusing the incoherent thing has been the most reliable compass I own.

So: I still can't pitch it in a sentence. Maybe that's the most honest thing about it. If the bet is right, the people this is for won't be sold it — they'll be shown it, by someone they trust, and they'll go open the box themselves.