Former OpenAI sales lead joins VC firm Acrew: OpenAI taught her where startups can build a ‘moat’

Alyssa Rosenthal, OpenAI’s first female sales leader, has found a new career: venture capital. She joins Acrew Capital as a general partner, working alongside co-founder Lauren Kolodny and the firm’s other partners, Rosenthal and Kolodny, per TechCrunch.

Rosenthal left OpenAI about eight months ago after spending three years in the AI ​​lab that saw the launch of DALL·E, ChatGPT, ChatGPT Enterprise, Sora, and other products. “I wasn’t initially looking to join a venture capital fund,” she told TechCrunch. “I was there to meet a lot of AI startups.”

But after growing OpenAI’s institutional sales team from two people to hundreds, she saw appeal when Kolodny pitched it to a venture capitalist. Instead of helping one startup with its go-to-market strategy, it can help a group of them.

During her time at OpenAI, she said, “I learned a lot about behavior, both on the part of buyers, and how people think about these purchases, and the gap between what most organizations think is possible and what they can actually deploy today.”

For example, it has first-hand insight into the kind of moat an AI startup can build that doesn’t leave it vulnerable when model makers like OpenAI launch competing products.

“Is OpenAI going to build everything and put every company out of business? You know, they’re doing a lot already: they’re in consumer, they’re in enterprise, they’re making a device. I don’t think they’ll go after every potential enterprise application,” she says.

So there is one dedicated moat for AI startups to offer specialisation.

Context is like a ditch

Additionally, she believes the key to a good startup moat will be “context” — or the information the AI ​​stores in its context window memory as it works on requests.

“Context is dynamic. It’s adaptive. It’s scalable. And I think what we’re seeing is moving beyond the basic sort of RAG toward the idea of ​​a context graph, which is ongoing,” she says, referring to Retrieval Augmented Generation (RAG), the de facto method from 2025 to reduce hallucinations by training LLMs on reliable, specific sources (which LLMs are required to cite).

There is still a lot of technology that needs to be developed in this area, from memory to thinking beyond pattern recognition.

“I expect real innovation here,” says Rosenthal. “I think this year we will see new approaches — the idea of ​​context and memory.”

But beyond startups that work directly on context engineering, Rosenthal believes that enterprise applications that integrate this engineering will have an advantage.

“Ultimately, when we talk about the moat, I think whoever owns and manages this context layer will become a huge advantage for AI products,” she says.

Another opportunity she sees: startups that don’t rely on the latest models from major labs, with their higher prices.

“I think there is room in the market for cheaper, lighter and innovative models in terms of inference costs,” she says. These models may not be at the top of the leaderboards for various criteria but they are “still very useful” and affordable.

“What I’m really excited about investing in is the applications layer,” she says. “I’m really interested in what will be the enduring applications built on all these different models, not just the core models.” It looks for startups with “interesting use cases” or using artificial intelligence to help an organization’s employees work more efficiently.

As for where they will find these startups, they will be working on their network among entry-level OpenAI graduates. Now that the AI ​​group is 10 years old, the alumni network has grown. Many of them have already founded startups that have generated big profits at high valuations, from OpenAI’s biggest competitor, Anthropic, to buzzy early-stage companies like Safe Superintelligence.

There is also a growing precedent for former high-profile OpenAI people becoming seed-stage investors. About a year ago, Peter Ding, former head of consumer products at OpenAI, joined Felicis. He’s been a fan ever since, and clearly enjoyed entering into big deals for hot startups like LMArena and Periodic Labs.

“I had a phone call with Peter a few months ago, and he helped me make the decision,” Rosenthal said of her choice to become an investor.

But Rosenthal may have a secret weapon for winning deals. It also has deep connections among enterprise AI users, which are the kind of buyers and beta testers these AI startups need.

Businesses still don’t understand how much AI can do for them. “There’s a really big gap and I’m very optimistic that we can fill it. It leaves a lot of green space for applications and companies.”