In February 2025 I wrote that in AI, the model is no longer the differentiator. Everyone has access to the same models. What separates startups is how good the product feels to use, and whether it spreads by itself. In July I pushed the second idea further: instead of asking how much money you can spend to buy users, ask how much engineering time you can spend to attract them.

I still believe both. But something I've seen in the fastest-growing companies I work with has changed how I think about the second half. The best ones don't just build distribution into their product. They build a separate product whose only job is distribution.

I call it the second product.

The first half of the thesis stays exactly where it was. The obsession with UX is for the first product, the one you sell. It's the thing customers touch, and when everyone has the same models, how it feels is most of what you have. Nothing in this essay is an excuse to make the first product worse. The second product is something you build in addition to it, not instead of it.

A Product Nobody Sees

The second product is exactly what the name says: a second, independent product the team builds for one purpose, to sell the first. (The emphasis is on "independent.") It isn't for customers. It isn't for the world. It's for internal use only.

The classic example is older than many of the founders reading this. In 1999, PayPal needed eBay sellers to accept PayPal. Sellers didn't care about payment methods. They cared about buyers. So PayPal's engineers built a bot. It emailed sellers to ask whether it could pay with PayPal, and if they said yes, it bought the item. The goods went to charity. [1]

Think about what that bot was. It wasn't a feature of PayPal; no PayPal user ever saw it. It was a separate piece of software, with its own logic and its own targets, and its only output was sellers who had met PayPal through a paying customer. It turned out to be a cheaper way to win sellers than advertising. That's a second product.

Note how it differs from a viral feature. A viral feature lives inside the first product and spreads when customers use it. The second product lives completely outside the first product and only exists to sell/distribute the first product.

Bots, Bids, Stores and Demos

Once you see the pattern, you see it everywhere. Usually years after it stopped being a secret.

The most common second product is an acquisition machine. Imagine a consumer app with a simple product: one screen, one button. Behind it, the team builds an engine that writes, launches, and tests hundreds of ads and campaigns a day, kills the losers, and feeds the winners. The engine can be orders of magnitude more complex than the app it sells. That's not a sign something has gone wrong. That's the point.

Booking.com is the grown-up version. Customers see a search box and a list of hotels. They never see the machine behind Booking's search ads: an internal system that processes over a billion events a day, runs machine learning models that generate bids, and pushes them to Google and Bing across thousands of ad accounts. Booking's engineers say it drives more than two million nights booked every day. Its first version was Perl scripts on a cron job, and it has been rewritten several times since. [2] The website is the first product. The bidding machine is the second, and it's at least as hard to build.

Cal AI is the version a teenager can build. It's a calorie-counting app: photograph your food, get the numbers. In January 2025, it quietly started more than a dozen brand-owned TikTok accounts, each with its own stream of content, and grew on that rather than on paid ads. [3] The app does one thing. The content machine behind it does a hundred.

Sometimes the second product is a way to be in many places at once. Before Wayfair was Wayfair, it was CSN Stores, a company almost nobody had heard of. It ran about 200 small websites, like racksandstands.com and cookware.com, each built to win one category in Google search. [4] Shoppers thought they were buying from small specialist shops. What the founders had actually built was a machine for spinning up stores that ranked. The furniture was the first product. The website production machine was the second.

And sometimes the second product is a demo. The best demo of a product is its output, delivered to someone who didn't ask for it. MeetGeek works this way from the inside: every summary it sends to a meeting's attendees is a demo. A second product can do the same thing from the outside, before there's a meeting at all. Imagine a cybersecurity startup whose second product maps the external attack surface of 200 companies a day, from public data only, and sends each CISO what it found. [5] By the time that CISO takes the call, the first product has already proven itself. The startup isn't pitching. It's showing a result.

Look how different these are: a bot that buys things, a bidding engine, a content machine, a store factory, a demo engine. The second product can be anything, as long as it's the embodiment of how the first product gets distributed. Ad optimization, demo distribution, a matching algorithm: whatever your distribution strategy is, turned into software.

The Invisible Moat

Here's what makes the second product different from everything else in growth: no one knows it exists.

A viral feature is public by definition. That's how it works. The moment it works, every competitor can see it, and today they can clone it in a week. A referral program, a share link, a multiplayer cursor: all visible, so all copyable.

The second product isn't public. It isn't advertised, it isn't on the website, it isn't in the launch post, and it isn't disclosed. It's for the team's eyes only. A competitor can see your app and copy your app. They can't copy a machine they don't know about. They don't even know there's something to copy. They see you growing and assume you raised more money, or got lucky, or have a better brand.

That's why the second product is the moat. Not because it's hard to build, though it usually is. Because it's invisible.

You could object that these things come out eventually. We know about PayPal's bot. We know about Wayfair's 200 stores. True. But notice when we found out: in books and case studies, years later. By the time the world learns about a second product, copying it is pointless. The company it built has already won.

Three Reasons to Build One

So why build one? There are three reasons, and any one of them would be enough.

The first is the one I just gave. Anything visible gets copied, and the cost of copying has collapsed. A feature that took a competitor six months to clone a few years ago now takes days, because AI coding agents do most of the work. If your distribution lives in a visible feature, you've handed it to your competitors. If it lives in a second product, you haven't.

The second is that most products can't be viral. I wrote a whole essay arguing for built-in virality, and when it's available, I still prefer it. But be honest about how often it's available. A meeting assistant is naturally viral, because meetings have other people in them. A cybersecurity tool isn't. Nobody shares their firewall. The same goes for software sold to labs, factories, and finance teams. For most B2B and deep tech companies, product-led growth in the strict sense was never an option. A second product is. Your product doesn't need to spread itself if you build another product that spreads it.

The third is that building two products is now cheap. Five years ago, telling a five-person startup to build two products would have been bad advice; you'd have split a tiny team in half. Not anymore. AI has made software so cheap to write that a small team can build what used to take a department. Most founders are spending that windfall on more features for the first product. The smart ones will spend part of it on a second product.

Built Like a Product

How do you build one? The answer is: literally as a second product.

That sounds obvious, but it's where most teams fail. They have the idea, someone builds a version of it in a sprint, it works for a while, and then it rots, because it belongs to nobody. It becomes a growth hack, and a growth hack is something you do once. A second product needs what any product needs: its own product owner, its own OKRs, its own roadmap. Someone has to take full responsibility for it in front of the CEO.

The reason isn't the amount of work. A second product, like a good viral feature, can be built once and keep working for years. The reason is focus. Every startup is tempted to spend all its engineering time on the first product, because that's what customers ask for and what investors look at. Without one person whose job is the second product, it will lose that fight every week. With one, it has someone fighting for it.

Until the company is big enough to hire for it, usually around the Series A, that person should be a founder. And at the very start, the second product shouldn't be software at all. Do it by hand first. Paul Graham's advice to do things that don't scale [6] still holds: send the first demos yourself, write the first hundred ads yourself, find the first sellers yourself. That's how you learn what works. The second product is what you build once you know: the machine that replaces your hands. Automate before you know what works, and you've built a fast way to do the wrong thing.

When the Second Becomes the First

One last thing. Sometimes the second product turns out to be worth more than the first. It's not a rule, but it happens.

Outreach started in 2011 as GroupTalent, a recruiting startup in Seattle. It was struggling, so the team built internal tools to automate their own outbound sales. Prospects noticed. They started telling the founders they didn't want the recruiting software; they wanted the sales engine. The company renamed itself after the second product and never went back.

AppLovin went further. In 2018 it began publishing its own mobile games, largely to generate first-party data for the machine learning behind its ad tools. The games were the first product; the ad engine was the second. By early 2025, the engine was bringing in more than three times the revenue of the games, and by mid-2025 AppLovin had sold the games business, ten studios, to keep only the engine. [7]

But this isn't a rule. PayPal's bot never became a business. It did its job and disappeared, and the payments company it fed sold to eBay for $1.5 billion. Most second products should end that way: invisible, effective, and in service of the first. If yours starts to look more valuable than the thing it sells, notice it. Just don't aim for it.

What you should aim for is simpler. Build a first product people love to use. Then build a second product nobody knows about, whose only job is to put the first in front of them.

If you build this, come talk to me at earlygame.vc

Notes

[1] Eric M. Jackson tells the story in The PayPal Wars (2004). Jackson was an early marketer at PayPal.

[2] Booking's engineers describe the system in an Apache Beam case study. The same page says Booking Holdings spent $5.99 billion on marketing in 2022.

[3] See this breakdown of Cal AI's accounts. Note what happened next: tools that run dozens of TikTok accounts for you are now sold to anyone. Once a second product becomes public, it becomes a commodity. By then the company that built it first already has the users.

[4] Fast Company and Xconomy covered the switch. The niche sites were folded into Wayfair.com in 2011 and 2012.

[5] The cybersecurity example is made up. The real ones aren't published, which is the point.

[6] Paul Graham, "Do Things that Don't Scale", July 2013.

[7] On the games as a data source, see Konvoy. In Q1 2025, AppLovin reported $1.16 billion in advertising revenue against $325 million from apps (CNBC). The sale to Tripledot Studios closed on June 30, 2025, for $400 million in cash plus a minority stake in Tripledot (PocketGamer.biz).