Do Things That Don't Scale

Most founders wait to be discovered. They build something, make it available, and trust that if the product is good, users will find it. But the startups that actually take off almost always required their founders to do something laborious, unsexy, and impossible to repeat at scale. This is a reading of Paul Graham's essay Do Things That Don't Scale.

The instinct to automate, to broadcast, to launch big is almost always wrong at the start. What works is the opposite: go get your first users one by one, make them irrationally happy, and stay in a deliberately small pond until you have enough heat to expand.

The most common unscalable thing founders have to do at the start is recruit users manually. You cannot wait for people to come to you — you have to go out and get them. Stripe is the perfect example. When someone agreed to try Stripe, the Collison brothers didn't send a link — they said "Right then, give me your laptop" and set them up on the spot. That's the Collison installation.

Founders resist manual recruiting for two reasons: shyness mixed with laziness, and the fact that absolute numbers look embarrassingly small at first. But 10% weekly growth on a hundred users means two million users in two years. The power is in the compounding, not the starting count.

Early startups are fragile. Investors, journalists, founders themselves — all make the mistake of judging them by established company standards. That's like looking at a newborn and concluding it could never accomplish anything. Airbnb survived because its founders spent thirty consecutive days going door to door in New York. Thirty days of unscalable effort made the difference between success and failure for what is now a hundred-billion-dollar company.

You should also take extraordinary measures not just to get users, but to make them ecstatically happy. Wufoo sent every new user a handwritten thank-you note. Being small is an advantage: Tim Cook can't send you a handwritten note after you buy a laptop. You can. The upper bound on customer experience is not set by industry convention — it is set by how attentive you are willing to be.

Sometimes the right unscalable move is to start in a deliberately narrow market — keep the fire contained until it gets really hot before adding more logs. Facebook started as a network just for Harvard students. The total addressable market was a few thousand people. But because it felt genuinely made for them, a critical mass signed up, and that density gave it the momentum to expand.

DoorDash is the public-market version of that same pattern: a messy, local, operationally heavy wedge that eventually became infrastructure at scale.

There is also a more extreme version: being the software yourself. When your user base is small enough, you can do by hand the things you plan to automate later. Stripe gave its first users what looked like instant merchant accounts — behind the scenes, the founders were signing them up manually with traditional processors. What founders fear, that manual processes will never automate, almost never happens. The far more common failure is building automation for a problem nobody actually has.

A startup idea is not a scalar. It is a vector: what you are going to build, plus the unscalable thing you are going to do initially to make it go. Most founders think hard about the first component and barely at all about the second.

The laborious, inconsequential-seeming actions of the early days are not a necessary evil to get through before the real work begins. If you recruit aggressively when you are small, you will still be aggressive when you are big. If you work hard to delight a handful of users, that habit embeds itself in the culture. For the best startups, the unscalable phase is the real work.