Will AI kill the venture capitalist?

The venture capital industry has always prided itself on being a business of judgment, instinct, and networks. It is, in Marc Andreessen’s words, one of the “last remaining fields that people are actually doing.” But if the recent surge in automation is any indication, the junior layers of venture capital - the training grounds where future Partners cut their teeth - may be the first to vanish. And what happens to a people-powered profession when there’s no pipeline of people?

Revolut founder Nikolay Storonsky doesn’t seem concerned. In 2023, he launched QuantumLight, a fully automated $250 million venture fund that replaces traditional sourcing and diligence with an in-house AI model named Aleph. According to Storonsky, Aleph allows QuantumLight to find promising businesses whilst avoiding a “crowds mentality” that often comes from human judgement. Aleph’s specialty? Spotting outlier founders - the black swans of venture investing - and doing it without a roomful of overworked associates. Storonsky is not alone. From Tribe Capital’s spinout of due diligence startup Termina to senior VCs building custom GPT workflows for screening and sourcing, the automation wave is rolling in fast.

Max Ruderman, founder of NLP-based sourcing platform Harmonic, recently noted in a convo with 20VC’s Harry Stebbings that the the product can automate much of the work of sourcing and diligencing, leaving VCs more time to ‘win the deal’. “You just don’t need to hire in the same capacity anymore,” says Grid Capital GP Jackie DiMonte. “Our hypothesis has completely reoriented.” But this raises an uncomfortable paradox: venture capital has long operated as an apprenticeship business. Junior hires slog through inbound, build mental models by reviewing hundreds of decks, and develop pattern recognition through lived exposure to thousands of decisions. If the analyst and associate ranks are hollowed out, what happens to that learning curve?

Andreessen’s answer is that venture, particularly early-stage venture, is fundamentally un-automatable. “There’s a taste aspect, the human relationship aspect, the psychology,” he argued on a recent podcast. His partner Ben Horowitz added that even if AI becomes better at picking, it still matters “who gets to pick” - access and trust, he suggested, cannot be automated. It’s a compelling argument - but one that hinges on the idea that today’s partners will always be around to pick. Without a bench of trained successors, venture risks becoming a priesthood of partners with no acolytes. What’s more, if junior talent never gets to compare enough companies, meet enough founders, and make enough mistakes, how can they ever develop the discretion Andreessen claims is so essential and human?

Oxford professor Thomas Hellmann identifies the tasks most susceptible to automation in VC: triaging deal flow, evaluating market comps, scraping data on team composition, and even initial diligence on IP portfolios. These are precisely the responsibilities currently assigned to junior staff. But Hellmann notes that AI struggles with “picking the winner” in a sparse data environment. “AI lends itself to portfolios of incremental innovation,” he writes. “For breakthrough innovation, it’s back to human judgment.” The underlying tension is that AI is fundamentally backward-looking. It finds patterns in existing data. But venture capital’s outliers - the true fund-returners - are, by definition, anomalous. As Keval Desai, managing partner at Shakti, puts it: “Early-stage VC is like picking Michael Jordan in kindergarten - when there isn’t much data available to feed into an AI model.” 

Even the poster children of data-driven investing have learned the limits. GV (formerly Google Ventures) famously built an algorithm to approve or reject investments based on market signals. It was quietly shelved in 2022. That said, AI has moved fast in three years. “The machine was far less sophisticated than today’s tools,” notes Desai. “But maybe something new will come along.”

The reality is we’re already seeing a bifurcation. Some firms, like QuantumLight and SignalFire, are going all-in on automation, aiming to track thousands of companies and maximize throughput. Others, especially early-stage firms, are doubling down on the human edge: founder trust, lived experience, and deep networks. If the first wave of automation wipes out junior analyst roles, it’s unclear who will fill step in to build relationships - can Partners atop an AI engine meaningfully build relationships at scale?

There’s also a legitimacy problem. If tomorrow’s GPs arrive with no sourcing or diligence background - no scars from passing on Uber or betting on Juicero - can they credibly claim the right to pick and the right to win? The bet may remain, but who’s earned the right to make it?

Will AI kill the venture capitalist? Not yet. But it will probably kill the venture apprenticeship model, and in doing so, it may strip the industry of the foundation it has relied upon for human discretion.

And to be clear, this definitely isn’t an argument against automating automatable tasks - but rather, questioning what happens to the apprentice model when cutting your teeth no longer adds to the critical workflow of the fund.

Find, select, win - learn by doing.

What advice do Partner-level VCs have for someone wanting to work in VC?

Don’t start at the bottom?