“Do things that don't scale” is outreach advice

The most-quoted startup essay ever written is, at its core, about outreach. “The most common unscalable thing founders have to do at the start,” Paul Graham wrote, “is to recruit users manually.” Every founder has read it. Almost none do it. The gap isn't conviction — it's that nobody talks about how brutal the actual work is.

What the famous examples actually did

• Airbnb flew to New York to meet hosts one by one — and photographed their apartments themselves.
• Stripe ran the "Collison installation": when a founder said "I'll try it," Patrick or John said "give me your laptop" and set it up on the spot.
• Pinterest's founder walked into Palo Alto coffee shops and asked strangers to try it.

Notice what these have in common: they're all direct, personal, manual recruitment of individual users. Not ads. Not launches. Not content. One person, going to another person, saying "I built this for you."

Why founders agree and still don't do it

Reading the essay takes eleven minutes. Doing what it says takes three hours a day, indefinitely: researching one prospect, finding a working email, writing something that isn't a template, logging the follow-up date, actually following up on it. Agreement is free; the labor is not.

So founders do the scalable-feeling things instead — a launch post, some content, a paid experiment — and six months later they have impressions instead of customers. The essay's point was never that manual recruitment is a nice-to-have. It's that at the start, it's the only thing that works, precisely because it doesn't scale: nobody else is willing to do it.

The spirit vs the mechanics

Here's the useful distinction. What made those stories work was the spirit: specific people, chosen deliberately, approached personally, followed up with relentlessly. The mechanics — copy-pasting addresses, checking a spreadsheet for who's on day 4 — contributed nothing. The prospect never knew whether the founder found their email by hand or not. They knew the message was about them.

That means the mechanics can be absorbed by software without losing what the essay was about, as long as three things stay human:

1. Choosing who. A narrow, deliberate ICP is a founder decision. (How to pick your first hundred.)
2. Approving what. Every claim that goes out under your name, reviewed by you.
3. Taking the conversation. The moment someone replies, it's you — that's the Collison installation moment, and it can't be delegated.

The modern version of the walk to the coffee shop

In 2013 the bottleneck was courage. In 2026 it's throughput: your buyers' inboxes are the coffee shop, and reaching them properly means research, verified emails, personal first lines, disciplined follow-up, and sending hygiene — for hundreds of people. That's the labor persist exists to absorb: you keep the choosing, the approving, and the conversations; the agent does the walking. The essay's advice survives intact — you're still recruiting users manually, you've just stopped being the spreadsheet.