The Cold Start Problem

On the paradox of beginnings and the fragile fight for ignition

This post is me thinking out loud. It’s a different taste than my previous ones, but lately I’ve been circling around one fragile, frustrating, and universal reality: every product begins in the cold.

Every product wants momentum. Every investor wants traction. Every founder wants proof. But before any of that exists, there’s only the beginning. No network effects. No users. No feedback loops. Just a product, a hope, and a blinking cursor.

Andrew Chen gave it a name: the cold start problem. But the dynamic predates tech.

Economists, sociologists, and system designers have all observed the paradox: a product isn’t valuable until people use it, yet people won’t use it until it’s valuable.

Most founders underestimate it. Many investors hand-wave it. But if you study failed startups (or the graveyards of products inside big tech), the pattern is consistent. They didn’t die because the idea was wrong. They died because ignition never happened.

Which leaves me with a handful of questions I can’t shake.

Why is building easy but ignition hard?

On paper, this should be the best time in history to start a company.

No-code platforms, open-source frameworks, and AI copilots collapse the cost of creation. A motivated engineer can build in a weekend what once took teams of ten and millions in funding. Infrastructure is elastic, distribution channels are global, and capital is sloshing around.

And yet, if you talk to founders, they’ll tell you the hardest part isn’t building anymore - it’s breaking through the noise. Distribution is the wall every new product smashes into.

The app stores that once promised open distribution have become infinite scrolls of sameness. Social feeds tilt toward incumbents who know how to game the algorithms. Even in SaaS, procurement budgets are sliced thinner than ever across hundreds of overlapping tools. Large enterprises now juggle 100+ apps, so every new entrant feels like another tax on attention.

This is the paradox of our era: building has never been cheaper, but starting has never been harder.

You can raise a billion dollars and still be stuck at zero if you can’t light the loop.

The cold start isn’t just a tactical hiccup anymore, but more so a bottleneck of modern product launches.

What happens when the loop never lights?

A failed ignition doesn’t always look like failure. Sometimes capital papers over the cracks.

Look at The Browser Company (I’ve written about it here before). The product won design praise and cultural attention, but the loop never truly lit: usage stayed flat, revenue near zero. Still, the company sold for $610M in cash. In consumer-facing categories, this kind of “failing upward” is possible if you attract enough capital and attention.

AI doesn’t grant that luxury.

Labs can raise billions, models can dominate benchmarks, and hype cycles can sustain the narrative. But without ignition (consistent, repeatable value delivered to end users), adoption stalls. And unlike consumer software, there’s no acquirer who can make unreliability valuable at scale. Reliability is the only thing that compounds.

Here’s the problem: reliability isn’t obvious from the outside. Benchmarks don’t capture it. Marketing can’t fake it. And investors can’t just assume it.

If ignition in AI depends on reliability, then the missing piece is a way to measure it. That’s where evals come in.

Evals are structured tests that measure how reliably an AI system performs on tasks that matter. They translate messy, unpredictable user conditions into quantifiable signals of trust. In a landscape where reliability is the moat, evals are the scaffolding that makes that moat visible and defensible.

What if the market itself is cold?

In Silicon Valley, the cold start is treated as a phase: wedge into a niche, find early adopters, and let compounding network effects do the rest.

But in emerging markets, the cold start isn’t a temporary hurdle. It’s the entire environment.

In the U.S., consumer apps can hack their way to 1,000 users overnight with a viral TikTok trend or a subreddit. Elsewhere, early adopters are scattered. They’re spread across languages, income levels, and platforms. Many aren’t online consistently. Others are online but wary, having seen apps vanish after a few updates or never localize properly.

Credibility, in that context, isn’t just branding. It becomes a feature. Customer support becomes your growth loop. Acquisition slows not because the product isn’t useful, but because you’re building on low-trust internet rails with fragmented payments, high data costs, and regulatory uncertainty. You’re not just selling value – you’re proving you’re not a scam.

Enterprise software doesn’t escape this dynamic either. Sales cycles stretch into quarters, sometimes years. Local firms run half their operations on Excel sheets and WhatsApp groups. Global incumbents still win tenders without localization, not because they’re better, but because “safe” beats “promising.” Unless you wedge very carefully, you end up competing with ghost incumbents — tools everyone uses, no one likes, and no one replaces.

In these markets, you’re not just solving the cold start for your product. You’re solving it for the system itself.

What actually works?

Despite the differences between AI agents, consumer apps, and emerging-market products, the ignition patterns converge.

Breakouts rarely arrive as sweeping revolutions. They usually wedge into narrow use cases, burn hot, and then expand outward. Facebook at Harvard. Airbnb in conference cities. Claude inside workflows.

The lesson: the smaller the spark, the hotter the fire.

Another constant is single-player value. If a product delivers value to one person on day one, it can sidestep the zero-user trap. Figma, Dropbox, and Calendly worked alone before network effects kicked in. In AI, coding copilots light up instantly because they solve a solo pain point. Browsing agents stall because they don’t.

Then there are high-energy users. The first customers aren’t passive adopters; they’re evangelists. Dense communities (college campuses, Discord servers, WhatsApp groups) turn one spark into ripples that spread faster than any paid campaign.

And finally, trust loops. In AI, adoption doesn’t just hinge on usefulness. It hinges on proof. Evals, compliance, localized support are not extras. They are ignition mechanics. They transform fragile beginnings into momentum.

These aren’t growth hacks. They’re the repeated physics of ignition.

What are the ignition strategies?

When founders hit the cold start wall, I’ve seen them reach for one of three plays. Each one is a different bet on how to light the loop.

The first move seems to be wedging into a tool.

The classic SaaS playbook has been: insert your AI feature into an existing workflow, charge per seat, and hope adoption compounds. It’s fast to get started, but inertia is the ceiling. Inside organizations, change management, retraining, and politics slow everything down. You get distribution, but you’re always fighting friction.

The second move is about acquiring distribution.

I mean skipping persuasion altogether. Buying the incumbents who already have users, then retrofitting them with AI. Inheriting the customer base, but also their baggage: messy P&Ls, cultural inertia, and the risk of overpaying for assets AI can’t actually fix. It’s a shortcut, but an expensive one.

The third move is building the operator yourself.

This is probably the boldest and riskiest path. Create an AI-native operator from scratch, owning the tech, the capital, and the operations under one roof. You control everything: the workflows, the customer touchpoints, the economics. But every execution mistake is yours, and there’s nowhere to hide.

These aren’t just business models. They’re ignition strategies. And the path you pick doesn’t just shape how you start. It shapes whether you survive.

So?

Cold starts demand more than vision, capital, or technical brilliance. They demand deliberate ignition.

Sometimes, as with The Browser Company, capital can disguise a failure to ignite. But in AI, there’s no such refuge. Reliability is the only way forward because reliability is what distributes, what builds proof, what creates trust.

Whether you wedge into a tool, acquire distribution, or build the operator yourself, the job is the same: earn your first believers, prove reliability, and close the trust loop.

Every great product has a moment before the magic when dashboards are empty, users nonexistent, and belief is the only fuel. Survive that, and the loop begins to turn. The system feeds itself. What once felt impossible becomes inevitable.

The cold start isn’t a hurdle on the way to your story. It is the story.