I wrote the first draft of this article in April and didn’t publish it.

The draft was about two kinds of company. The first is the consumer brand built by someone who already has an audience. Kim Kardashian launched Skims in 2019, and last November it raised money at a $5 billion valuation. MrBeast launched a chocolate brand, Feastables, in 2022. By 2024 it was bringing in about $250 million in sales and more than $20 million in profit, while his media business lost almost $80 million on similar sales.

The second is software built by a very small team. In 2025 a developer in Israel named Maor Shlomo built an AI app builder called Base44, mostly on his own and without outside funding. Six months after he started, Wix bought it for $80 million in cash.

What connects them is the cost of getting started. For decades, building a consumer brand usually took a TV budget, and building software usually took a big engineering team. Both costs have fallen a long way, and I expected markets to split into many more, much smaller companies. I called it the great fragmentation.

I still expect that. What has changed since April is my view of what it takes for one of those companies to last. I wrote then that distribution and speed were replacing capital as the main advantage. I now think a company also needs an advantage that’s hard to copy.

Brands built on an audience

Marc Andreessen made a point in late 2023 that I agree with. Brands like Coca-Cola and Kraft Mac & Cheese exist, he said, “because of the media of the era in which those brands were created.” Television let a company build an emotional connection with millions of people at once. He called the result an “unnatural configuration,” a person feeling attached to a corporation. With a creator, “it’s a relationship with a person.”

That’s why I expect more creator brands. A creator already has the audience and its trust, so the brand can start without a TV budget.

Software built by small teams

The cost of building software fell too. In March 2025, Y Combinator’s CEO, Garry Tan, said that startups no longer need a team of 50 or 100 engineers, and that for about a quarter of YC’s winter batch, 95% of the code was written by AI.

If you’re one of those engineers, that’s worrying. It’s a valid concern, I share it, and I don’t have the answers.

You can see the shift in who starts companies. Carta found that 36% of the startups founded on its platform in 2025 had a single founder, about double the share of a decade ago.

I see it at Trellace too. We’re a small team, and we now build a lot of our own internal tools with AI. We build too much, and I’m the worst offender. I think it’s easy to build something that’s okay. To build something excellent is still hard. To build something that doesn’t degrade over time, because it isn’t slop on the inside, is also still hard. Then there’s maintaining everything once it’s built, which I completely underestimated.

Where the money went

When it costs less to build a company, founders need less money to get started, and more of what they raise goes to distribution. Venture money, meanwhile, has been concentrating for a few years, and I expected it to. A huge amount of money came into venture funds during COVID, and valuations went up with it. When those valuations didn’t carry into later rounds, the investors who back venture funds got more careful, and raising a fund got harder.

In the US, venture funds raised $222 billion across 1,817 funds in 2022 and $75 billion across 907 funds in 2025. In the first half of 2026, three firms took in almost half of everything raised. The investments concentrated too. Over the same six months, OpenAI and Anthropic took $217 billion of the $510 billion invested worldwide.

What I find interesting is how much the largest firms now do for their founders besides investing. They’ve built platform teams, staff whose job is to help the firm’s companies with things like hiring and reaching customers. A 2023 study of 850 venture firms found that about half had a platform team, double the share in 2000, and that more than 90% of the firms managing over $1 billion had one. Andreessen Horowitz has more than 600 employees, and Cursor’s COO has said the firm’s platform team introduced Cursor to nearly 200 CTOs at target customers in their first year working together.

I think there are two reasons. When the bar to start a company is this low, distribution is critical, so help with reaching customers is worth a lot. And firms are making fewer, larger bets, often on the same companies, so the founders who are doing well get more attention when they raise and can pick among offers. They compare what each firm can provide around the money. That has always been true, and it’s especially true now.

What I’d change from April

I still think distribution and being first to market matter more than they used to. But competitors can build as cheaply as you did, and that includes much larger companies.

Cursor is a good example. The AI coding tool passed $1 billion in annualized revenue about 32 months after it launched. But Anthropic and OpenAI sell coding tools too. According to spending data from Ramp, Cursor’s share of that market fell from 41% in June 2025 to about 26% in May 2026, when Anthropic had half of it. In June 2026, SpaceX, which had taken over Elon Musk’s AI company xAI in February, agreed to buy Cursor for $60 billion. The deal closed in August, and Cursor said joining SpaceX gives its team access to the computing power to train stronger models.

That’s a great outcome for the founders and their investors. Growing fast got Cursor a long way, and competitors still took share from it.

What I look for now

When I look at a company, I start with how good the founder is at selling. Founders have to sell the product first. They have to sell talented people on the vision so they can hire the best. And they have to sell investors on it so they can raise enough money at a competitive valuation.

The second thing I look at comes from Hamilton Helmer’s book 7 Powers, which I think is the best framework I’ve seen for competitive advantage. Helmer describes seven sources of lasting advantage, among them scale, network effects, switching costs and a resource that competitors can’t get. I know opinions differ on moats, and one argument is that things move too fast now for a moat to matter. I think being able to adapt is almost as important as having a moat, and I still look for one.

The advantage I think is critical right now is proprietary data, or a proprietary method for getting insights out of data that anyone can access. The cost of general intelligence and of writing code keeps falling. Sam Altman has written that the cost to use a given level of AI falls about 10x every 12 months. Every company gets that cheaper capability, and a company with its own data or its own method can apply it to something competitors can’t easily copy.

So I still expect many more, much smaller companies, because starting one keeps getting cheaper. The ones I expect to last have a founder who can sell and an advantage that’s hard to copy.