Unlock the Editor’s Digest for free
Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.
Whose job will be next for the AI chop? Some kinds of work are already squarely in the crosshairs as artificial intelligence flexes its muscle, from software coders to call centre workers and junior lawyers. But a recent study raises the possibility that AI could also bite the hand that feeds it: that of the venture capitalist.
Researchers at Oxford university and venture firm Vela Partners asked a range of large language models, including those from OpenAI and Google, to spot technology founders with big-money potential. The LLMs were shown anonymised data on 9,000 would-be Mark Zuckerbergs from LinkedIn and Crunchbase; a successful founder was defined as one who would later raise at least $500mn, or sell a company for that amount.
Pitted against the VCs, the AI models came out on top. For context, top firms broadly get it right one time in 20. But some LLMs did much better. OpenAI’s GPT-5 correctly picked a winner half the time. China’s DeepSeek-V3 beat the lot, able to single out a future success story six times out of 10. It was also able to do so more cheaply than most other models.
While these findings might weaken the cult of the VC, they do the opposite for the founders themselves. The AI study implies that indeed, it’s the individual and their record that predict future success, more than funding, timing or luck. That will be music to the ears of Silicon Valley, which has gone all-in on the idea of the talented visionary.
True, it’s not an either-or. Venture firms are already using AI to enhance their operations, from portfolio analysis to streamlining communications with their own investors. Andreessen Horowitz, a backer of AI giant Anthropic, is one that touts the benefits of AI-powered offices. Only a complacent VC wouldn’t be using AI to screen potential investments.
Where AI might not excel is in taking risks. Identifying whether a founder will thrive is different from identifying all future winners — qualities known respectively as precision and recall. DeepSeek was so conservative it missed most of those destined for greatness. Google’s Gemini found more of those with the golden touch, but picked lots of duds in the process.
Both precision and recall matter in venture capital, an industry where just a few investments drive most of the profit. No doubt, AI will get better at both. It remains to be seen, though, whether a technology that uses past data to predict the future can identify real outliers. When it comes to spotting mega-profitable oddballs, humans are likely to retain their edge.