Crunchbase: VCs demand defensible moats

Venture capital is relearning an old lesson in a new AI wrapper. Building software got cheaper and faster, but that did not make investing easier. It made differentiation harder. The news here is that Crunchbase’s May 6 essay put that shift in plain language: if everyone can ship product with the same foundation models, investors have to look past raw technical chops and ask what actually keeps a startup from being copied. ### Why are VCs talking about moats again? Because AI changed the cost curve. A tiny team can now prototype features that used to need a real engineering org, which means “we built it fast” is less impressive than it was even a couple of years ago. Crunchbase’s piece says technical expertise still matters, but it no longer differentiates on its own when the product layer is easier to reproduce. ### What changed for early-stage teams? The bar moved from building to owning. Crunchbase points to Carta data showing the average seed-stage company now has just over 6 employees, down from more than 10 in 2021. Lean teams are doing more with less, but that also means investors expect more proof per dollar — sharper founder-market fit, clearer go-to-market plans, and some reason the company gets stronger as customers use it. ### So what no longer counts as a moat? Model access by itself, basically. If a startup’s main pitch is “we use the latest model to do X,” that edge can disappear as soon as a better model ships or a competitor plugs the same API into a similar workflow. The hard truth is that foundation models are drifting toward infrastructure. Great for building around generic AI SaaS. ### What does count now? Things that are painful to copy. Proprietary data is one. Deep workflow integration is another. Distribution, trust, compliance, and switching costs matter more too. Insight Partners makes the same point from the investor side: the defensible layer is often workflow depth plus data that compounds as customers use the product. ### Where does SeedLegals fit in? SeedLegals pushes a similar idea from the founder side. Its AI-first startup playbook argues that the winning companies are not just bolting AI onto an old product. They redesign the product and the operating workflow around what AI is actually good at. That sounds subtle, but it is a big difference — an AI feature can be copied, while an AI-native workflow can become the way work gets done inside a customer’s team. ### Why does workflow matter so much? Because habits are sticky. If your software sits in the middle of how a recruiter screens candidates, how a law firm drafts documents, or how a clinic handles intake, ripping it out is expensive even if a rival has similar model quality. Workflow depth is like plumbing — nobody brags about it, but once it is inside the walls, replacing it is messy. That is the kind of moat investors can underwrite. ### Does this mean technical founders matter less? No — but the job changed. Investors still want strong builders. They just also want founders who control some scarce advantage outside the model itself: customer access, domain knowledge, unique data, regulated-market know-how, or unusual distribution. In other words, the best AI startup team is no longer “the smartest model people.” That is the shift Crunchbase is pointing at. ### Bottom line? AI lowered the cost of making software, but it raised the standard for funding it. For founders, the message is simple: don’t pitch the magic trick. Pitch the thing that gets stronger after the trick becomes common.