Why the Best Founders Do Things That Don't Scale
Emrah · Airbnb's founders photographed apartments by hand. Stripe's founders installed their own product on strangers' laptops. Neither looks like a scalable business. Both were building one.
The Story
In the early days of Airbnb, the founders flew to New York and went door to door. Not to pitch investors. To take better photos of the apartments already listed on the site. The listings were ugly, the photos were dark and badly framed, and nobody was booking them. So Brian Chesky and his co-founders picked up a camera and did the work themselves, one apartment at a time.
Stripe's founders, Patrick and John Collison, did something similar a few years later. When a new user signed up, they didn't send a link to a setup guide. They asked for the person's laptop, sat down next to them, and installed the payment integration on the spot. People inside the company started calling it the "Collison install." It became a small piece of company folklore, and a very serious lesson underneath the joke.
Paul Graham wrote about this pattern in a 2013 essay called Do Things that Don't Scale. His argument is almost a contradiction on its face: the startups that go on to build massive, scalable companies almost always start by doing something that scales terribly. Manually. Slowly. One customer at a time.
The Idea
Most founders assume that building a "real" company means building for scale from day one. Automated onboarding. Self-serve signup. A funnel that runs itself while you sleep. Graham's point is that this instinct, applied too early, is usually a mistake.
In the earliest stage, the product is not good enough yet to sell itself, and the market is not proven enough to justify investing in infrastructure for growth that hasn't shown up. What you have instead is a small number of real users and an open question: does this actually solve their problem? Manual effort is how you answer that question fast. You watch someone use your product in real time. You hear the sentence they say right before they get confused. You notice the step they skip, the field they misread, the moment their face changes. No analytics dashboard gives you that.
There's also a second, quieter benefit. Doing the unscalable thing lets you deliver a level of service that a company ten times your size could never match. A personal thank-you note. A live install. A founder who answers support tickets in real time. Big companies can't do this because it doesn't scale. Early on, that's exactly the point. It's not a permanent strategy — Graham is clear that growth eventually needs to come from something repeatable — but the manual phase is the bridge that gets you to a product worth automating.
How Founders Should Read This
For a founder, the discomfort here is usually pride, not logic. There's a specific kind of embarrassment in explaining to a friend, or worse, to an investor, that you spent last week hand-writing thank-you notes to your first fifty users. It doesn't feel like founding a company. It feels like busywork.
Kevin Hale, who co-founded Wufoo before it was acquired by SurveyMonkey, leaned into exactly this instinct. His team sent a handwritten card to every new signup. It didn't scale past a few hundred users. It also turned those first few hundred users into people who told their friends. The mechanism wasn't the card itself — it was the signal the card sent about how much the founders cared about getting this right.
The trap most founders fall into is treating the manual phase as something to escape as fast as possible. They build the automated version before they've learned what's actually worth automating, and end up scaling a version of the product that skips the parts that made it work. A better way to think about it: manual effort now is how you find out what the automated version should even look like. If there's a part of your onboarding you're still doing by hand three months in, don't be embarrassed. Ask whether you actually understand it well enough yet to hand it to a machine.
How Investors Should Read This
An investor looking at early traction has a different question in mind: is this growth real, or is it a founder working eighteen-hour days to manufacture a metric? The answer matters more than it might seem, because it predicts what happens after the check clears.
Here's the useful reframe. Manual growth, done well, is not a weaker signal than paid growth — it's often a stronger one. A founder who can describe exactly how they won their first hundred users, what almost made each of them churn, and why most of them are still active six months later, is showing you something a growth-marketing dashboard cannot: a working theory of the customer, built from direct contact. A founder whose only answer to "how did you get your first users" is "we ran ads" hasn't necessarily done anything wrong, but they also haven't necessarily learned anything yet.
The red flag isn't manual effort. It's the absence of it. If a team scaled to thousands of users purely through paid acquisition before anyone on the founding team had personally onboarded a customer, that's worth a harder look at retention numbers specifically, because rented attention and earned attention behave very differently once the money runs out.
Fund math sharpens this further. A venture portfolio depends on the handful of companies that compound retention over years, not the ones that produce an impressive but shallow signup number in month two. Manual-era retention — the percentage of hand-won users still active a year later — is one of the closest things to real proof of that compounding a very early check can get.
Where the Two Views Collide
Here's where it gets interesting, and where most conversations about this idea stop short.
Founders often hide the manual period rather than showcase it. There's a quiet shame attached to saying "we're still doing this by hand," as if it signals amateurism instead of discipline. So founders reach for the story that sounds more impressive: bigger numbers, an automated funnel, a growth chart that looks inevitable. Ironically, this is precisely backwards. What should feel embarrassing isn't the hand-built period — it's never having gone through one at all.
Investors, meanwhile, have their own blind spot. The manual phase is a strong signal only within a window. If it stretches on for two or three years with no credible path to a repeatable, scalable motion, it stops being founder discipline and starts being evidence that no scalable model exists yet — or that one might never exist. The skill an investor needs here is distinguishing "this founder is patiently learning what to automate" from "this founder has found a service business, not a startup," and that distinction usually only shows up when you ask directly: what have you learned from doing this manually that you're now ready to systematize?
The person who understands both sides of this tension has an edge at the table. They read early manual effort as evidence, not excuse. And they know exactly when to start asking the harder question about what comes next.
The Takeaway
If you're building something right now, here's a concrete way to apply this today: look at your own product or the last pitch you evaluated, and ask whether the early growth came from something the founder built with their own hands, or from something they simply paid for. The answer tells you more than almost any other single data point available at that stage.
Source: Paul Graham, “Do Things That Don’t Scale” (2013) — https://paulgraham.com/ds.html
