Nabeel Hyatt

AI is pushing venture from spreadsheet-solvable puzzles back toward mysteries requiring first-principles judgment. Nabeel Hyatt argues that no one knows what a model will do next week, so firms optimized for SaaS metrics, coverage, and consensus risk becoming obsolete. The host worries that spreadsheet-driven investing could make firms “dinosaurs.” Winning firms will be small, curious, and comfortable making subjective bets amid fog of war.

Venture’s organizational incentives increasingly reward markups rather than enduring outcomes. A principal “is not actually waiting for an exit. They just want a promotion, man,” so the rational move is learning what the next-stage fund wants, investing one month earlier, and packaging a four-month markup. That system is especially fragile when nobody can predict what will be hot nine months later.

Revenue velocity is no longer a sufficient proxy for AI-company quality. Products can reach $10 million ARR within months and still “probably be dead in 2 years”; the investable question is whether the founders possess enduring taste, exceptional execution speed, and the capacity to reinvent themselves. “This industry is all about exceptions,” making checklist investing structurally mismatched to the asset class.

Product matters chiefly as evidence of how founders think. Hyatt does not infer investability from a polished interface; he interrogates the decisions embodied in “the thing that comes out of their hands” to distinguish real executors from hucksters. The strongest AI founders combine speed with the taste to discard acceptable metrics and rebuild when a product “doesn’t feel right.”

Price is a test of conviction, but excess capital can alter—and kill—the company being underwritten. Hyatt is not a value investor, yet a $25 million round may create a fundamentally worse business than a $10 million round by changing hiring and execution before the company is ready. Deployment-focused investors therefore cannot assume the same eventual $5 billion outcome after forcing in more cash: “That company will probably raise another hundred million and then they might just die.”

The attractive AI opportunities sit beyond adaptation, in evolution and especially revolution. Hyatt separates incumbents adding AI, products creating genuinely new workflows, and businesses possible only because the technology exists; Spark avoids mere adaptation and seeks behaviors that “sear into your brain.” Much of today’s incubator output is instead a thin arbitrage engineered to show roughly 10% weekly growth before a seed round.

Anthropic’s bull case is a full-stack learning loop, not an indefinitely protected model moat. Owning the customer interface creates usage insight and “data exhaust” that can improve both product and model; Hyatt sees this as the “wisdom of experts,” exemplified by Descript learning from every expert edit. DeepSeek does not overturn his thesis because Spark never believed capital expenditure alone was the barrier to model competition.

Vertical agents become durable only when they attack hard, decade-long problems and have second, third, and fourth acts. Labor replacement can enlarge previously uninteresting niches, but an agent that perfectly solves today’s workflow may soon face 25 clones and drive the market toward zero. Hyatt’s filter is blunt: “Pick a job that’s hard”—ideally one where a hallucinating MVP earns initial demand while leaving years of innovation ahead.

🔗 Original source & video: Nabeel Hyatt, GP @ Spark Capital: To Win in AI, Investors Need to Change Their Approach | E1255