For two decades, software was the moat. Build a better product, win the market. That moat has shrunk - code is faster, cheaper, and easier to write than ever. "Better software" alone no longer delivers a venture return.

What's replacing it: workflow ownership and proprietary data flywheels.

Two distinctions matter most.

Closed-loop vs. Open-loop:
→ Closed-loop products control the workflow end-to-end. They accumulate trust, compound data, and lock in switching costs.
→ Open-loop products generate outputs but stall at human judgment. They struggle to improve and increasingly resemble features.

Real data moats vs. claimed ones:
"Data is the new oil" is directionally correct but misleading in practice. Most "data moats" in startup decks aren't moats at all. The data that matters is generated within the workflow, is domain-specific, and improves the product with every use.

Harvey and Glean are the clearest examples - narrow workflows that produce data that compounds.

3 Question Moat Test:
1.     Does the product own the workflow?
2.     Does it operate in a closed loop?
3.     Does the data compound with every use?

If not, it's probably a feature, not a company, and vulnerable to death by LLM.

What's the strongest moat you've seen built in the last 12 months - and does it pass these 3 tests?

This is the second post in a series on rebuilding the VC playbook. 

Last week: why it's fracturing. 

Next in the series: why incumbents should not be underestimated.

The AI Moat Matrix
Where a company sits on these axes of workflow ownership and data flywheel strength determines its outcome ceiling.

The best companies will have both.

A sneak peak into the conversations that will be happening in San Francisco at VC Unlocked: AI Edition 2026 (June 1-4). 

VC Unlocked: AI Edition is a four-day immersive venture capital program designed to equip experienced fund managers, VCs, and angel investors with the tools, insight, and conviction needed to lead in the next wave of AI investing.