TL;DR
Most products don’t fail because they’re bad. They fail because they launched into emptiness. Andrew Chen calls this the Cold Start Problem, and it’s the silent killer of otherwise great products. The fix isn’t scale. It’s finding the smallest possible network where your product becomes undeniably useful, then nailing that.
1. The trap most founders walk into
You build something genuinely better than what’s out there. You launch. And nothing happens.
Not because people don’t want it. Because your product is only valuable when other people are already using it. That’s a network effect, and most founders only think about it as the upside. The flywheel. The thing that kicks in once you’re big.
But network effects cut both ways. Before they help you, they actively work against you. Andrew Chen calls this anti-network effects. New users show up, the network is empty, they leave. The product looks broken, even when it isn’t.
Some examples of how this looks in practice:
Slack is incredible when your team is on it. It’s useless if you’re the only one.
Airbnb with one listing in one city is worthless. There’s nothing to book.
Uber with too few drivers is frustrating. You open the app, no cars, you delete it.
Wikipedia with no articles is just a blank page asking you to write one.
The value doesn’t exist until the network exists. And the network doesn’t form on its own. This is what kills most products that should have worked. Not the idea, not the team, not the engineering. Just emptiness on launch day.
2. The cold start is especially hard in AI
If you’re building in AI right now, this trap is bigger than ever. The model is mostly the same across products. The wrapper is mostly the same across products. What’s left to differentiate on is the data, the workflow, and the network of people using it.
Most AI products today have some networked component, even if founders don’t see it that way. Your fine-tuning improves with usage. Your context gets richer when more teammates feed it. Your output quality compounds when more users correct it. Your distribution often depends on someone telling someone else.
Which means the cold start applies to you. The first ten users determine whether the next hundred ever show up. If those first ten don’t hit a moment where the product feels alive, the data flywheel never starts, the word of mouth never starts, and the model never gets the feedback it needs to actually be better than what’s on the shelf.
3. How the best companies engineered around it
The companies that won didn’t wait for the network to form. They forced it. Often through methods that looked unscalable, embarrassing, or borderline cheating at the time.
Airbnb manually posted listings to Craigslist to find their first hosts. Their early team also flew to New York and photographed apartments themselves to lift listing quality.
Uber launched city by city. They subsidized drivers heavily, ate the losses, and didn’t move to the next market until the current one had average wait times under five minutes.
LinkedIn seeded with Silicon Valley tech workers first. A small, tight, professionally connected group where one signup actually meant five more would follow.
Tinder threw lavish parties at USC. To get in, you had to download the app. They engineered a dense atomic network on one campus before opening anywhere else.
Snapchat designed the product so that one user had to pull in another. You can’t snap to no one.
Bank of America launched the very first credit card by mailing it to 60,000 residents of Fresno, California in 1958. Both sides of the network, cardholders and merchants, in one small city. Then they expanded.
Different industries, different decades, same playbook. Find the smallest group where the product actually works. Saturate it until it’s undeniable. Then copy and paste.
4. The atomic network: smaller than you think
The atomic network is the smallest possible group where your product is genuinely useful. Not interesting. Not promising. Useful enough that people would be upset if you took it away.
Most founders dramatically overestimate how big this group needs to be. Here’s what it actually looked like for products that went on to win:
Notice how small these are. Two people. Ten people. Five hundred. The number you need to make it work is almost always smaller than the number you think you need.
Density beats total size. A thousand random users scattered across the world is a worse start than fifty users sitting next to each other on the same campus, in the same office, or in the same Slack workspace.
The Founder Launchpad
Our founding cohort started last week and the first session was unbelievably amazing - this cohort includes incredibly talented founders building in a diverse set of areas.
We had 60 applicants from which we chose 7 for our first cohort. We’ll be starting our second cohort early June - register now to be a part of it.
5. The hard side is where the work is
Every network has an easy side and a hard side. The hard side is the smaller group of people who do most of the work and create most of the value. They’re also the people you have to work hardest to get.
On Uber, drivers are the hard side. Without them, the app is empty.
On Airbnb, hosts are the hard side. Without listings, there’s nothing to rent.
On YouTube, creators are the hard side. About 1% of users make 99% of the content people actually watch.
On Wikipedia, prolific editors are the hard side. A tiny group writes most of the encyclopedia.
For an AI product, the hard side is often whoever trains the system. The teammate who corrects the output, tags the data, or builds the workflow that everyone else benefits from.
If you don’t know who your hard side is, that’s the first question to answer. Once you know, your job is to make their experience disproportionately good. Pay them, feature them, build tools just for them, do whatever it takes to keep them engaged. The easy side will follow on its own.
6. Scale is the wrong question early on
Most founders start with the wrong question. They ask, how do we get to a million users? It tells you nothing about what to do tomorrow morning.
The right question is much smaller and much harder:
Find the smallest network where this becomes undeniably useful. Nail that one. Make sure those users would actively complain if it disappeared. Then build the next atomic network adjacent to the first. Then the next. That’s how networks actually grow.
The companies that became enormous didn’t start by trying to be enormous. They started by being completely indispensable to a tiny group, and then they did it again, and again, and again.
The question worth sitting with
What is the smallest network where your product becomes undeniably useful? Not nice to have. Not interesting. Undeniable. If you can’t answer that, you don’t have a launch plan, you just have hope. And hope is not a strategy.
Build your network this week
Define your atomic network. Get specific. Not “startup founders.” Not “marketing teams.” What is the smallest possible group, in one place, at one time, where your product would feel alive? Write it down. If it’s bigger than a few hundred people, you haven’t gone small enough.
Identify your hard side. Who are the small group of users who create most of the value for everyone else? Founders, power users, content creators, top sellers, expert reviewers. Build a list of names. Real people. Then design your launch around making their lives unreasonably good.
Do something that doesn’t scale. The early playbook is manual. Onboard the first ten users yourself. Hand-match the first connections. Sit on calls. Edit the first pieces of content. Anything that gets the network to density faster is worth the time.
Stop tracking vanity metrics. Total signups don’t mean the network is working. Density does. Repeat use does. The right question is, does this network sustain itself for one week without you pushing it? If yes, you have an atomic network. If no, you have a list.
Resist the urge to expand. The biggest mistake after a small win is going broad too fast. Saturate before you scale. One city before five. One campus before ten. One company before a hundred.
The cold start problem doesn’t go away because you ignore it. It just kills you quietly while you’re busy planning for scale. Solve the small problem first. The big one takes care of itself.
