Last quarter we evaluated 47 GenAI startups. We invested in 3. Here's a transparent look at why we passed on the other 18 that made it to partner discussion.

Pattern 1: GPT wrappers without distribution

8 of the 18 were thin wrappers on OpenAI/Anthropic APIs. The product worked. The demos were impressive. But there was no path to defensibility — anyone could replicate the wrapper in a weekend. We pass on these unless there's a clear distribution advantage that compensates.

Pattern 2: Market that's already winning

5 deals targeted markets where incumbents were already shipping AI features (e.g., AI-powered code completion against Copilot). We're cautious about Day 2 entrants — unless the incumbent has a structural blind spot, late entry rarely wins.

Pattern 3: Founder couldn't articulate the wedge

3 deals had great founders and large markets, but couldn't crisply explain their entry wedge. 'We'll start broad' is not a wedge. We invest in clarity, not optimism.

Pattern 4: Unit economics that don't pencil

2 deals had compelling products but inference costs that ate 60%+ of revenue at scale. Margins don't fix themselves. If the math only works at 100x the volume, that's a bet we can't underwrite.

What made the 3 we invested in different? Each had a structural advantage we couldn't see being replicated — proprietary data, a regulatory moat, or a distribution edge that compounded. Plus founders who could articulate exactly why now and why them.