Earlier this month I was in Las Vegas for HumanX, a new AI industry conference, alongside about 3,000 other attendees. I also moderated a panel on-stage with early-stage investors from three VC firms that have overwhelmingly been placing their bets lately on artificial intelligence: Kleiner Perkins, Bain Capital Ventures and Primary Venture Partners.
For these three firms, investments in AI-related companies reached an astonishing 80% to 90% of their new portfolio deals in 2024.
We discussed how AI is changing the terms of early-stage investing and chatted about the risks and rewards of investing in AI at the nascent stages. What follows are highlights from my conversation on stage with Kleiner's Leigh Marie Braswell, Bain's Aaref Hilaly, and Cassie Young of Primary. (You can also watch the whole conversation below.)
AI has been "less predictable and straightforward than the past waves," said Braswell of Kleiner.
However, "the demand from consumers, from enterprises, for this technology and the amount of real proof points of the intelligence paradigm shift that's happening is just unprecedented," she said.
Braswell said she has seen product-market fit with multiple startups going after opportunities in code-generation, AI developer tooling, customer support enabled with AI, sales, legal and medical scribes.
Kleiner Perkins is an investor in AI coding startup Codeium, AI corporate search engine Glean, sales assistant Nooks and Ambience Healthcare, an AI assistant for healthcare clinicians.
The question for investors is: "What are the nonobvious applications?" asked Hilaly of Bain Capital Ventures. "There's still a huge number of those, mainly because the capabilities of models continue to advance. And so you have this shifting forward frontier, particularly around what's called reasoning and the model's ability to fully automate things and perform tasks."
Bain Capital Ventures is an early investor in edtech MagicSchool AI, customer support startup Decagon, and Moveworks, a platform that provides IT service automation and was recently acquired by ServiceNow. AI advancements unlocked opportunities for all of those companies.
In an increasingly competitive startup environment, "I love to ask founders -- what is your head start?" said Braswell.
Founders have to assume that their competition is rational and smart, Braswell said. So, founders have to ask themselves: "How long would it take, and how far could you be in front of them before they start to copy you?"
Hilaly agreed: "There are no moats, there is no defensibility, certainly at early-stage."
In that world, startups compete by bringing products to market faster and continually, he said, "because customers and users will recognize that."
Bain Capital Ventures also looks for founders that have "an intuition for their end user," he said.
Even when companies do see early adoption, that might not be sustainable over time.
"Traction is a huge red herring right now," said Young of Primary Venture Partners. That further unsettles venture funding.
Young cited a recent report that as much as 60% of generative AI purchases are coming from innovation and experimental budgets.
"I see churn red lights everywhere," she said. "So even the head start makes me nervous."
Primary Ventures is an investor in supply chain company Lyric, 1Mind, which makes photoreal avatars for go to market teams and Alma, a mental health startup.
We have a term for AI founders: "next-level missionaries," Young said.
"I love that framing because I think it underscores the importance of customer obsession in the category they are building in," as well as needing to win in crowded spaces, she said.
AI companies need less capital because you can do more with smaller teams, said Hilaly, and given the amount of capital available, it makes investors' lives harder. Companies often "can grow revenue more aggressively early in the company's life," he said.
Agentic AI was a recurring theme throughout the three-day conference.
"There are definitely people on Twitter or X that will tell you, 'agents are here, and they're coming for your jobs. And you can hand off any task to an agent today.' That's wrong," Braswell said. "Agents don't work in terms of, let's automate every single end-to-end task."
AI tools are great at building software workflows. Braswell cited a portfolio company that used Windsurf from Codeium, a portfolio company, to build its own sales software tooling, and canceled contracts that cost hundreds of thousands of dollars.
"Eventually the models will get so good at predicting your workflow and your personalized workflow that you won't want to switch from one application to another, because this one knows you really well," said Braswell.
"Models will be good enough to really figure out what to do and take a series of actions, at which point everything we think of as a software application basically becomes an agent, because why would it not?" said Hilaly.
