The first ten are hand-sold

No automated funnel replaces personally messaging the first ten buyers — what gets missed when the checkout page gets built before anyone confirms they want to click it.

By this point a project on the shelf has a price, a named smallest paid version, and a payment path written down — a processor named, the entity question at least asked, and the line between what’s free and what costs money drawn. Written down, not wired: none of it has been built yet, and none of it has to be before the next step. The instinct most people reach for next is a funnel: a landing page, an email sequence, maybe some ad spend. That is the wrong tool for the number you need first.

Ten customers do not need a machine built to find them. They need you, personally, going and finding ten people.

Why you do not build a funnel for ten customers

A funnel is built for volume — it assumes enough traffic arriving that a conversion rate means something, and it only earns its cost once the labor of building it is smaller than the labor of reaching people one at a time. At ten customers that math runs backwards. Writing ad copy, testing landing-page headlines, wiring up an email sequence all take real hours, and every one of those hours could instead go toward finding an actual person who might actually buy.

Paul Graham names this directly. Recruiting a startup’s first users by hand, he argues, is not a failure to find a better channel — it is close to universal, something nearly every startup ends up doing, and founders who wait for users to arrive on their own tend to wait a long time. His own instruction is blunt: “You can’t wait for users to come to you” (Paul Graham, Do Things that Don’t Scale). The essay also endorses starting narrow on purpose — Facebook launched to Harvard students only, not to the internet at large — which is the same move as picking one project’s first ten rather than trying to sell all seven at once to everyone who might conceivably want any of them.

What hand-selling actually looks like for a solo builder

The essay’s sharpest image for manual recruiting is what Graham calls the Collison installation, a technique he credits to the Collison brothers (the founders who went on to build Stripe) and says became a term used at YC once other founders started copying it. The unambitious version of hand-selling is offering to email someone a signup link. The Collison version, on hearing someone say yes, is asking for their laptop right there and setting the whole thing up before the conversation ends.

For a solo builder that scales down to something concrete: get someone on a call instead of sending a link, and walk them through checkout while you are both looking at the same screen. Send the invoice yourself and follow up the next day if it is unpaid, instead of waiting for an automated cart-abandonment email to fire on a cart that only ever had one item in it. The point is not theater — it is removing every step between “yes” and “paid” that a stranger could quietly fail to finish alone.

Graham backs the idea with two more examples worth borrowing. Airbnb’s founders went door to door in New York to recruit hosts and personally improve their listings. Wufoo sent a handwritten thank-you note to every new user for the company’s first few years. And he offers a line worth hearing before any of this starts feeling like too much effort for ten people: he says he has never seen a startup hurt by trying too hard to make its initial users happy.

Where people get burned

The essay treats consulting — doing custom, one-off work for a customer — as the canonical example of something that doesn’t scale, and warns it turns dangerous the moment you start charging for it, because a founder paid to build custom things has quietly started a consultancy, not a startup. That caveat cuts directly against a common instinct on a solo shelf: using paid bespoke work as the on-ramp to a product, by building whatever the first paying customer happens to ask for. There is nothing wrong with that outcome for a solo builder — a paid custom-build practice can be a perfectly good business. It is simply not the same business as the product sitting on the shelf, and Graham’s warning is about noticing the switch, not about avoiding paid work outright.

What the first ten tell you that analytics cannot

An analytics dashboard reports a bounce rate. It cannot tell you which sentence on the page made someone hesitate, whether they actually reached for a card or just said “maybe later” to be polite, or which feature they assumed was included and got quietly annoyed to find wasn’t. Selling to ten people by hand produces exactly that kind of information, because you are there for the hesitation.

This is the same category of evidence the shelf’s demand-evidence field already asks for — just gathered live instead of reconstructed after the fact from whatever a dashboard happened to log. A stranger saying, out loud, “I’d pay for that if it did X” is worth more than a week of pageviews, because it arrives with the reason already attached.

Retrieval check

A stranger agrees to buy on a call, then goes quiet for three days after you send the invoice. What does a funnel tell you about why? What does hand-selling let you find out that a funnel structurally cannot?

Check your answer

A funnel gives you a drop-off point and nothing else — the metric says the invoice step lost a customer, not why. Hand-selling lets you follow up as a person and just ask, and the real answer is usually one of a small number of specific things: the price landed wrong once it was in writing, the person never actually had authority to pay, or the invoice step itself was confusing. Each of those changes what you do next in a way “cart abandonment: 1” never will.

When to stop hand-selling

Hand-selling assumes there is somewhere to hand-sell to, and a solo builder starts this exercise without the scaffolding that makes it easy for a startup founder. Graham’s founders are recruiting inside a demo day audience, an investor network, and — for a B2B startup — hundreds of fellow YC companies who make an instant first market. None of that exists for a project shipped alone in the evenings. The first ten have to come from somewhere real: people you already know who have the exact problem, users of a different project on your own shelf who might plausibly want this one too, a specific forum, Discord, or mailing list where people already discuss this problem, a local group that meets in person. Specific and findable, not a demographic.

Aside

Read past the tactic and the essay is arguing toward something much bigger than ten happy customers. Graham’s own arithmetic: a startup with 100 users needing 10 more the following week is only 10% growth, and 10% a week compounds to roughly 14,000 users in a year and about 2 million in two — manual recruiting is how you bootstrap the start of that curve, not the destination. He also describes gradually switching to less manual methods as the numbers climb, which makes “doesn’t scale” a phase he expects a founder to exit on purpose, not a permanent way of operating. A project that stops happily at ten paying customers is not the outcome the essay is arguing toward at all. This lesson keeps the tactic — go find your first buyers yourself — and drops the growth obligation it was built to serve. That substitution is this course’s choice, not something the essay itself endorses.

Hands on

Name where ten real buyers already are

Done when: A new First ten field in PORTFOLIO.md, added under the project furthest along, lists ten specific people or places — each one specific enough to act on this week, and none of them “post on social media.”

Pick the project on the shelf you’d call furthest along by your own verdict — the one with something real to sell right now, not the one still missing basic screens.
List ten specific people or places where a plausible buyer for that project already spends time. A person means someone you could name: a former coworker who complained about this exact problem, a specific friend of a friend. A place means somewhere addressable: a named subreddit, a named Discord server, a named local meetup, a mailing list you could actually email. “Instagram” and “Twitter” are not places by this definition — they are broadcast channels, and broadcasting is exactly what this exercise is not.
Next to each one, write the specific action, not the category: the DM you would send, the exact person you would call, the post you would write inside that one community and nowhere else.
Cross off anything that is secretly a broadcast wearing a disguise — a scheduled post to a public feed, an unaddressed message dropped into an entire server. If it does not have one buyer’s name or one room’s name attached, it does not count as one of the ten.
Add a First ten field to that project’s entry in PORTFOLIO.md — it does not exist there yet — and write the finished list of ten into it.
Bring the list into the chat. I’ll push hardest on any entry that is really “post on social media” wearing a disguise, since that is the one that always tries to sneak back in.

None of this is free, either. Finding and closing ten people by hand costs real hours, and those hours deserve the same measured-not-guessed treatment as any other cost already sitting on the shelf. Once the ten have told you what you needed — that the price holds up once it is real money, that the pitch lands in words you actually used rather than words you guessed at, that a stranger who owes you nothing will still pay — hand-selling has done its job for this round. What comes after that is a different problem: how to reach an eleventh customer without personally being on every call.

What this does not cover

Hand-selling gets a project its first believers. It says nothing about whether anyone can find the project at all once those ten are done — a live URL sitting with zero visitors is not a launch, it is just inventory with a working checkout attached. The gap between shipping something and anyone seeing it — the distribution gap — is what comes next.