Sam Lessin on VC and AI

Q2 2025: VC and markets are disaggregating. AI changes everything and is a bad investment.

Sam Lessin shared his views on the VC market, seeing how his 2023 predictions held up. These stood out to me:

1. "AI" is quite practically reorganizing financial markets.
   Inbound is dead (it all looks the same / return to feudal based system vs. open one), software is no longer investable, SaaS is worthless, acquisition logic has shifted, and startup capital consumption patterns have changed. AI's impact on the capital markets / building ecosystem might be more profound than the tech itself (similar to cloud's outsized impact on markets, but on steroids).

2. AI ‘cherry on top’ businesses.
   The best businesses are ones which are great on their own, would have been great pre-AI but have a non-obvious AI ‘cherry on top’ where the curve of LLMs, etc. make them more profitable, more useful, etc. in critically non-obvious ways. IF it is too obvious, the market is too efficient /overpriced.

3. AI ‘second order effect’
   Society will change a lot because of AI… what people value, how culture works, how the ‘adversarial’ information and content spaces function – the second order effect (what do we need now to defend against AI, because of AI) is fertile ground.

The broader theme of the document is that the VC factory line model, where entrepreneurs followed a basic SaaS model with legible targets, is over, and now a company’s stage depends on which segment of investors is interested in it. AI seed investing has been a washout for most, as the rapid evolution of underlying tools makes it hard to build and LLM wrappers aren’t good businesses.

Stepping back, I think Sam does not go far enough on how transformative gen AI will be.

First, Inbound is dead because it’s become too easy to spam things. This has implications for every kind of content, not just financial market structure. Second, while AI can help existing strong businesses, the way they disrupt cost structure means that existing strong businesses are not that strong, because they are so operationally inefficient. Third, the second order effects he is concerned about seem to focus on safety but we are already in a poorly managed “adversarial” information that is not trending towards a healthy equilibrium.

The whole deck is thought provoking, and recommended.