Jenny Xiao came to venture from research. Following a PhD at Columbia, she landed an early seat at OpenAI during the run-up to ChatGPT, and had a front-row view of a venture industry that, in her telling, was still funding NFTs while the actual technological shifts were happening in labs nobody was paying attention to.
Jenny is now a Partner at Leonis Capital, a research-driven fund she co-founded straight out of OpenAI to fix the problem she kept running into: most VCs don’t fully understand the technology they’re investing in.
In this episode of Going Solo, we discussed:
The best AI products always arrive ahead of the model. Lovable started as GPT-Engineer years before Sonnet 3.5 made vibe coding actually work. Jenny only backs founders who can see 12-18 months down the model capability curve, because the ones building for today’s models are usually building something a bigger lab commoditises within a year.
Leonis’ algorithm is better than they are at picking winners. Leonis backtested its internal founder-scoring system against 20,000 companies, and it beat the partners. Everything the fund knows (CRM data, memos, judgment) lives in a single GitHub repo so an AI agent can query it instead of asking a human where a five-year-old file went.
“Founder first” has become “founder only”. Jenny argues that plenty of investors now use “I just back great people” as cover for not understanding anything else about a market. Her question: how do you even prove you’re a good judge of character?
Most “neo labs” are the wrong shape for venture money entirely. Leonis mapped 119 of them across capital needs and speed to revenue. Only 8% fit the traditional venture model; 60% need OpenAI-sized capital without OpenAI-speed monetisation, which is a much better fit for sovereign wealth than a ten-year VC fund.
We also got into why open-source models are closing the gap on closed-source ones faster than anyone expected, why she thinks Anthropic’s current lead is more fragile than the market is pricing in, and her view that the next ten years of AI is less about assistance and more about software making decisions on our behalf.
Hope you enjoy it.
DG
