How to Do Things That Don't Scale Like Airbnb

Trying to look big on day one is a common startup mistake. The contrarian advice, from Paul Graham and Brian Chesky alike, is to do things that don't scale: hand-serve a tiny market until those users love the product, then figure out how to industrialize it.

What Does "Do Things That Don't Scale" Actually Mean?

Doing things that don't scale means solving the product-validation problem by hand for a tiny group of users before investing in growth. The advice comes from Paul Graham's 2013 essay Do Things That Don't Scale, written by the co-founder of Y Combinator, and it rests on one line he repeated to founders: it is better to have 100 people who love your product than 1 million who kind of like it.

The reasoning is statistical as much as strategic. A representative sample of about 100 people, drawn carefully from a well-defined target audience, is enough to tell you whether you have found something a much larger market would want. Once that signal exists, you can attack the separate problem of scaling. Founders who start with scalability usually get neither: they build infrastructure for a product nobody loves yet.

Why Did Airbnb Wait 16 Years for a Second Product?

Brian Chesky, co-founder and CEO of Airbnb, described the trap in a widely shared 2025 interview: Airbnb had roughly $100 billion in annual gross bookings, about 0.1% of global economic output flowing through the platform, and still only one successful product. For around 16 to 18 years, nothing the company launched after its core home-rental business worked.

His diagnosis was that Airbnb carried the burden of global scale from day one of every new initiative. When the company launched Airbnb Experiences, its feature for booking tours and activities, it rolled out in about 100 cities at once and it did not work initially. The lesson he drew: Airbnb itself started in only New York, Uber started in San Francisco, and new products should start in one market and expand only after that market works.

Chesky's stated model after that correction is staged: run one pilot city, and if it works expand to ten, and only then industrialize. By the time of the interview he reported running 10 to 20 such pilots, with a target of 50 to 70, and said the second and third successful businesses took about 16 years to arrive.

How Does the 100-User Sample Size Math Work?

The number 100 is not arbitrary, and the video that inspired this article walks through it with a sample size calculator. For a population of 8 billion, a 95% confidence level and a 5% margin of error require about 385 respondents. Relax the margin of error to 10%, which is reasonable for entrepreneurial decisions where being roughly right beats being precisely wrong, and the required sample drops to about 97.

The useful part is that the sample size barely moves with population size. Whether your addressable market is 100,000 people or 1 billion, roughly 97 to 100 well-chosen respondents are enough at those settings. In a niche of only 100 companies, you would need to talk to most of them; in a market of 500,000 doctors, about 97 conversations carry the same evidential weight.

Two caveats keep this honest. First, this is an analogy, not a proof: your 100 users must genuinely represent the target segment, the same way election polls depend on proper sampling. Second, confidence level is the chance your sample method holds up across repeated draws, while margin of error is the range around the result. Entrepreneurship tolerates error because a wrong early call is cheap; a late one is not.

What Did Airbnb's Apartment Photographer Prove?

The canonical example is Airbnb sending professional photographers to shoot the first New York listings. Sending a photographer to every host clearly does not scale, and that was the point. The founders were testing a specific hypothesis: better listing photos would increase conversions and bookings.

It worked, and the discovery reframed the scaling question. The question stopped being "can we afford photographers for every listing?" and became "how do we get photography-quality listings at scale?" That second question is solvable with tooling, incentives, or host education. Without the artisanal first step, the company might never have learned that listing quality was the growth lever.

The same pattern now applies to software products, Chesky argues, including AI products: prototype hand-crafted for a small group, iterate until those users love it, then manufacture. Scaling is industrialization, and industrialization is a different problem from finding product-market fit.

How Do You Split a Startup Into Artisanal and Industrial Phases?

The practical method is a two-phase process. Phase one is artisanal: pick one city, niche, or customer subset small enough that you can talk to everyone in it, and do unscalable work without worrying about unit economics until the model is proven.

Phase two only begins once roughly 100 target users love the product. Then the question becomes how to industrialize what worked. The contrast Peter Thiel draws in Zero to One, a book by the PayPal co-founder, supports the same sequence: a monopoly in a small market beats a small share of a large market, because large markets are crowded and small markets can be grown.

The failure mode is skipping phase one. If every early decision is filtered through "will this scale to 1,000 customers?", you will refuse the photographer, never learn what users love, and scale a product nobody wants. Trying to warm the ocean with a hairdryer, as the metaphor in the source video puts it, tells you nothing; warming one bathtub tells you everything.

Frequently Asked Questions

• Who wrote "Do Things That Don't Scale"? Paul Graham, co-founder of Y Combinator, published the essay in 2013. It collects advice he gave startups at YC, including the line about 100 people who love you versus 1 million who sort of like you.

• Where did the "100 people who love you" advice come from? Graham attributes it to Paul Buchheit, the Y Combinator partner who created Gmail. The Gmail story attached to it: Google's launch bar was getting about 100 people inside the company to love the product, which reportedly took roughly two years.

• Does 100 loving users really predict millions? Not mathematically, but as a sampling analogy it holds: at 95% confidence with a 10% margin of error, about 97 respondents suffice regardless of whether the population is 100,000 or 1 billion. The sample must genuinely represent the target segment.

• What was Airbnb's most famous non-scalable act? Sending professional photographers to shoot the first New York listings. It proved that listing photo quality drove bookings, and only afterwards did the company solve how to achieve that quality at scale.

• Should every startup avoid early scaling? The advice targets validation-stage startups. Once product-market fit exists in a pilot market, staged expansion and industrialization are exactly what Chesky recommends: one city, then ten, then industrialize.

Turn Your Own Video Into a Written Playbook

This article exists because a 16-minute YouTube video about non-scalable growth contained a lesson worth reading, not just watching. If you have the same kind of knowledge sitting in your own videos, whether it is a founder interview, a strategy breakdown, or a lesson like this one, Skala Blog turns it into a written article. Paste the YouTube URL, let it transcribe the video, and generate a structured, publishable draft you can edit and ship.

The same two-phase logic applies to content: prove the idea in one format by hand, then scale it. Skala Blog handles the unscalable first step for you.

Source video