Every week we meet 30+ founders claiming their startup has a defensible moat. Most are wrong. Foundation models are commoditizing. Open-source is closing the gap on closed-source within months. UX patterns are copied within weeks. So what actually creates lasting defensibility for AI-first companies?

1. Proprietary data flywheels

The strongest moat we see is a data flywheel where customer usage generates training data that improves the product, which attracts more customers. The key word is 'proprietary' — public web data is not a moat. Customer interaction data, expert-labeled domain data, and behavioral signals that competitors can't access — that's the moat.

2. Workflow lock-in

If your product lives inside the customer's daily workflow — synced with their CRM, embedded in their docs, integrated with their ticketing — switching costs become enormous. This is why vertical SaaS AI tools beat horizontal copilots. Specificity wins.

3. Distribution moats

Sometimes the moat isn't the tech — it's that you can acquire customers cheaper than anyone else. Could be a brand, a community, a content engine, or a partnership. We've backed companies whose primary edge is distribution, and they've outperformed technically superior competitors.

What doesn't count as a moat

First-mover advantage in AI? Not a moat. Better model accuracy than competitors today? Not a moat — that erodes in 6 months. Founder pedigree? Helpful but not a moat. A great team without a structural advantage will struggle once well-funded competitors enter.

When we evaluate AI investments, we're asking one question: in 3 years, what makes this company impossible to displace? If the answer involves only the technology, we usually pass.