The new moats are the old moats: 3 lessons from Jerry Chen

I’ll admit something. Every time a new AI tool launches, a small voice in my head asks whether anything I’m building is actually defensible. If someone can copy a product in a weekend, what’s left?

That’s why Jerry Chen’s essay The New New Moats stuck with me. Jerry is a partner at Greylock, and in 2023 he revisited a framework he’d written six years earlier on how startups build defensible businesses. He openly called out where his old predictions were wrong, which I loved. Here’s what I took from it.

1. The value is in joining things up

Jerry splits software into three layers. At the bottom are systems of record, where a company keeps its core data on customers, employees and assets. At the top are systems of engagement, the screens people actually use, from web to mobile to chat. In the middle sit what he calls systems of intelligence: products that pull data from several of those records and turn it into something useful.

His point is that the middle is where startups can win. Build on one company’s data and that company can simply copy you. Combine several sources and you’re much harder to replace. The more people use the product, the more it learns, which makes switching away painful.

The example that made this click for me is Glean, which Jerry lists among the AI apps of this wave. It doesn’t own a company’s data. It plugs into the tools where that data already lives, like Slack, Google Drive and Jira, so you can search across all of them at once. Slack can’t easily copy that, because Slack only sees Slack.

2. Clever tech on its own is a shallow moat

This one stung a little. Jerry argues that technology without a real customer problem around it gets squeezed “between open source and a cloud place.” Startups that were a thin wrapper on ChatGPT ended up competing with the model provider itself. His point is that nobody cares what database runs their food delivery app, as long as dinner turns up on time. Customers buy the complete solution, not the clever part.

3. The new moats are the old moats

This is the bit I keep coming back to. Even if anyone can build with AI, what makes a business durable hasn’t changed: workflow, brand and trust, network effects, scale and go-to-market. Jerry says you need product go-to-market fit, not just product-market fit, and that AI doesn’t change how startups market, sell or partner.

He also notes how fast brand strength evaporates once customers lose trust. That applies to every business, not just AI ones.

What I’m taking away

I don’t know which tools will matter in two years, and Jerry is honest that he doesn’t fully know either. But I know where to spend my energy: earning trust, getting distribution right and building something customers would miss.

Thanks to Jerry Chen for writing it. The full essay is on the Greylock blog.