Answers are grounded in this paper and cite their sources.
Artificial intelligence (AI) agents can now carry out much of the first-pass work of venture capital due diligence. This article presents a framework for doing so in seed transactions, with three propositions that later work can test, and applies it as a proof of concept to 161 seed rounds announced in 2025, using three models of different size and cost. The agents produce a structured first review of a deal in seconds and for cents. Because today’s models may already know how past deals turned out, the author hides each company’s identity, limits every comparison to rounds announced before the deal, and measures what the models still know. Across models, the agents ask for the same missing information, chiefly the terms of the round and evidence on the investors. The midsize and largest models rank the deals in nearly the same order, as closely as two runs of the same model do; the smallest model ranks them differently. The author argues that any lasting advantage is more likely to come from combining the agents with a firm’s own information and analytical methods than it is from the agents alone; the author also separates findings about the 2025 seed market from artifacts of the sample and the design.
No takes yet. Share an insight, caveat, or question.
Jitesh Gurav (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: