Why early startups should do things that don't scale

Executive overview

Startups obsessed with scalability from day one build products no one wants. Paul Graham's 2013 essay inverted the prevailing VC wisdom: ignore scalability until you have users who force you to care about it.

The goal of doing unscalable things is to maximise learning before you build the wrong thing. Manual, hands-on work with early customers gives you knowledge no algorithm can generate.

Do the thing that doesn't scale until you've answered the question worth asking — then, and only then, invest in scaling.

Why scalability became the wrong obsession

• Google's dominance made scalability a cultural fixation — founders and investors demanded "scalable" everything from day one.
• VCs would not fund a company that couldn't articulate a scalable business model.
• The real startup-killer is building something no one wants, not an architecture that can't handle load.
• PG's essay was a direct inversion: stop worrying about later problems before you've solved the first one.
• The "Field of Dreams" startup — perfect infrastructure, zero users — became the cautionary archetype.

The Airbnb and Algolia examples

• Airbnb had no traction; PG told them to go photograph listings themselves rather than wait for hosts to upload quality photos.
• That manual effort turned the flywheel — skipping it would have killed the company before it started.
• Nothing is beneath the early founder. Cold emails, admin, cold calls — the best founders dive in regardless of their prior seniority.
• Algolia implemented their own search directly inside Product Hunt's GitHub, giving the founder access and submitting pull requests themselves.
• That hands-on setup created a depth of relationship — candid feedback, real context — that no self-serve onboarding could replicate.
• The core question: "How much can I learn today?" Doing the job manually before building software ensures you're building the right thing.

Fleek: learning a marketplace by hand

• Fleek (W22) connects secondhand clothing wholesalers with retail shops.
• With no website and no inventory, founders approached London wholesalers, took boxes of clothes for six hours, sold them to shops by hand, and returned with cash.
• Repeated over four months, this revealed what sells, pricing elasticity, timing, and demand patterns — knowledge impossible to extract from an empty marketplace page.
• Once those lessons were in their bones, they built the platform to replicate the same matching at scale.

Instacart: launching without a single partnership

• Conventional approach would have meant corporate negotiations with Trader Joe's — potentially years of legal and procurement overhead.
• Instead, Instacart used YC funding to buy one of every item at Trader Joe's, photographed everything in a studio over a weekend, and launched with a full catalogue on Monday.
• By the time Trader Joe's noticed, Instacart had enough customers to be a desirable partner rather than a nuisance to shut down.
• The unscalable shortcut gave them the negotiating leverage a cold pitch never would have.

DoorDash: one-afternoon MVP

• The first DoorDash product used Google Drive for menus, an HTML/CSS page, a Google Form for orders, and Find My Friends for real-time dispatch.
• One founder picked up the order; another tracked via Find My Friends and texted ETAs manually.
• A team of Stanford engineers and MBAs could have built the "fancy" version — they chose not to, because the hard question was demand, not technology.
• Startup advantage: do the thing that would get you fired at a big company. Big companies cannot compete with founders willing to embrace errors and chaos.
• Turn on all the water and fix the cracked pipes as they appear — don't spend two years checking every pipe first.

Founder FaceTime as a competitive weapon

• Early founders can give out personal cell numbers and promise hands-on support that no corporate account manager ever will.
• Customers are not buying the product at this stage — they are making a bet on the founder.
• A scrappy early-stage founder beats a well-funded competitor at the relationship layer every time.
• Founder proximity to customers also makes recruiting easier — the same urgency and commitment applies.

When to stop doing unscalable things

• "Do things that don't scale" has a natural end: when you've answered the key question, you have licence to invest in scaling.
• Failure mode: getting addicted to consulting revenue — it answers the question but won't grow 10x.
• Optimizely manually ran A/B tests as a consulting service to prove willingness-to-pay, then built software to make themselves more productive at that service, then scaled the software.
• A useful signal: can you grow 10x in a year? A consultancy cannot; a real software business can.
• Good advisors and investors help founders recognise when to flip the switch — founders inside the work often can't see it themselves.
• The unscalable phase also preserves optionality: no hard-coded software means you can pivot the next day without technical debt blocking you.