Sam Lessin vs Seth Rosenberg: Is AI a great startup opportunity or will incumbents capture all the value?

Summary

Sam Lessin vs. Seth Rosenberg: AI as startup opportunity

The central question is whether AI primarily enriches incumbents or creates room for new companies to build durable businesses. Lessin and Rosenberg have been sparring on this publicly for roughly a year — including a literal billboard war on Highway 101 — and the debate reflects a genuine fault line in how venture capital is thinking about the current cycle.

Lessin's position: AI is an incumbent accelerant

Lessin argues AI is about as far from a disruptive innovation as you can get. His framing is that mobile and cloud were also largely sustaining — the same companies got bigger — and AI is an even cleaner example of that pattern. The winners he'd have bet on two years ago are the same ones already sitting at the top: Meta, Nvidia, the hyperscalers. His actual AI investing strategy, stated directly, is "thank God I held on to some Facebook stock."

His core critique of the startup opportunity isn't that AI isn't powerful — it's that the AI layer itself is commodity. Whether a company uses OpenAI, Anthropic, or something else barely matters, because the switching cost approaches zero and every competitor has access to the same tools. That dynamic levels everyone up rather than creating a wedge.

He's also skeptical of the "AI replaces labor" playbook. Using accountants as an example, he argues many professional roles exist primarily to shift liability, not to execute tasks. An AI that makes fewer errors than a human accountant still loses to the human, because what you're really hiring is someone to absorb blame when the IRS comes calling. The AI commoditizes the task but doesn't displace the trust relationship.

The one category Lessin does believe in he calls "AI cherry on top" businesses — companies that are fundamentally sound and happen to get more efficient because of AI. His own example is Merit First, a skills-based hiring platform he helped start with Joe Lonsdale. The pitch is that AI finally makes test-based hiring viable at scale: it can help design meaningful evaluations, deter cheating, and grade thousands of applications in ways that were previously impossible. But he's explicit that you're investing in a new approach to hiring, not in an AI company.

Rosenberg's position: new markets, new networks

Rosenberg doesn't dispute that AI won't kill incumbents. His argument is narrower: AI opens up new markets that didn't exist before, and a few of those will produce outlier outcomes worth funding.

He frames it around two buckets. The first is new content types enabling new networks — analogous to how the smartphone camera didn't just improve photography apps but eventually produced Instagram, Snap, and TikTok. The equivalent question for AI is what new types of content can now be created by everyone. He names games ("what's the next Roblox"), AI-generated music (he points to Suno), and software creation as candidates. The bet is that a platform built around a genuinely new creation type could bootstrap a closed ecosystem before distribution gravitates back to existing pipes.

The second bucket is AI displacing labor in large, historically human-dominated markets — legal, compliance, recruiting. He cites Greenlight, a portfolio company, as an example: large banks and fintechs employ thousands of compliance analysts doing checkbox work, and software is starting to absorb that. His framing is that AI is competing with BPOs and labor spend, not with incumbent software vendors.

He does concede Lessin's point that outbound sales automation may be self-defeating. Once every outbound email is AI-customized, no one can tell what's real, and the entire channel degrades. He raises Tome — a slide and sales automation company in Greylock's portfolio — as a case where the honest question is whether it can survive Salesforce bundling comparable features into its existing CRM. His answer is that it doesn't need to beat Salesforce to succeed, just find a segment that wants a more modern architecture.

Where they land

Both agree the middle is the worst place to be. Lessin puts it directly: being a subscale company competing with hyperscalers using venture capital is a recipe for disaster, because the hyperscalers have orders-of-magnitude more money and time. Rosenberg's version of the same point is that the barbell logic of AI favors either hyperscalers or small, highly specialized operators — ideally individuals with deep domain expertise and genuine community trust. He takes the "one-person billion-dollar company" thesis seriously.

The disagreement that remains is probabilistic. Lessin thinks the chance of building a durable AI-native startup is low enough that almost all pitches deserve a pass. Rosenberg thinks the outlier upside — even at low probability — justifies selective bets, particularly on founders with domain depth who are using AI to rearchitect a market they already understand rather than pitching AI as the product itself. Both agree, with some irony, that the concept of an "AI company" will probably be meaningless within a few years. Everything is just a company that happens to use the tools available.