OpenAI’s former sales leader joins VC firm Acrew: OpenAI taught her where startups can build a ‘moat’ 

OpenAI’s archetypal income leader, Aliisa Rosenthal, has recovered a caller career: task capital. She’s joining Acrew Capital arsenic a wide partner, moving alongside founding spouse Lauren Kolodny and the firm’s different partners, Rosenthal and Kolodny archer TechCrunch.

Rosenthal near OpenAI astir 8 months agone aft a three-year sprint astatine the AI laboratory that saw the motorboat of DALL·E, ChatGPT, ChatGPT Enterprise, Sora, and different products. “I wasn’t initially looking to articulation a VC fund,” she told TechCrunch. “I was retired determination gathering with tons of AI startups.” 

But aft increasing OpenAI’s endeavor income squad from 2 radical to hundreds, she saw the entreaty erstwhile Kolodny pitched her connected task capital. Instead of helping 1 startup with its go-to-market strategy, she could assistance a portfolio of them.

In her clip astatine OpenAI, “I learned a batch astir behavior, some connected the broadside of the buyers, however radical are reasoning astir these purchases, and the spread betwixt what astir organizations deliberation is imaginable and what they tin really deploy today,” she said.

For instance, she has firsthand penetration into what benignant of moat an AI startup tin physique that won’t permission it susceptible erstwhile exemplary makers similar OpenAI launching competing products.  

Will OpenAI “just physique everything and enactment each institution retired of business? You know, they are doing a batch already: they’re successful consumer, they’re successful enterprise, they’re gathering a device. I don’t deliberation they are going to spell aft each imaginable endeavor application,” she says. 

So 1 moat is for endeavor AI startups to connection specialization. 

Context arsenic moat

Additionally, she thinks the cardinal to a bully startup moat volition beryllium “context” — oregon the accusation the AI stores successful its discourse model representation arsenic it works connected requests.  

“Context is dynamic. It’s adaptable. It’s scalable. And I deliberation what we’re seeing is going beyond benignant of the basal RAG towards this thought of a discourse graph, which is persistent,” she says referring to Retrieval-Augmented Generation (RAG) the de facto method arsenic of 2025 to minimize hallucinations by grooming LLMs connected trusted, circumstantial sources (and having the LLM mention them). 

There’s inactive a batch of tech that needs to beryllium developed for this area, though, from representation to reasoning beyond signifier recognition. 

“I expect existent innovation here. I deliberation this twelvemonth we volition spot caller approaches — the thought of discourse and memory,” Rosenthal says. 

But beyond startups moving straight connected discourse engineering, Rosenthal thinks endeavor apps that cook it successful volition person the advantage. 

“Ultimately, erstwhile we speech astir moat, I deliberation who owns and manages this discourse furniture volition go a ample vantage for AI products,” she says. 

Another accidental she sees: startups not gathering atop a large lab’s state-of-the-art models, with their precocious prices.  

“I deliberation determination is country successful the marketplace for cheaper models that are lighter value and innovate connected inference costs,” she says. These are models that are not, perhaps, astatine the apical of the leaderboards of assorted benchmarks but “are inactive precise useful” and much affordable. 

“Where I’m truly excited to put is connected the exertion layer. I’m truly funny successful what volition beryllium the durable applications built connected each of these antithetic models, not conscionable connected the foundational models,” she says. She’s seeking startups with “interesting usage cases” oregon that usage AI to assistance endeavor employees enactment much efficiently.  

As for wherever she’s going to find these startups, she’ll beryllium moving her web among OpenAI’s alums for starters. Now that the AI outfit is 10 years old, the alums web has grown. Many person already founded startups that person raised large bucks astatine precocious valuations, ranging from OpenAI’s biggest competitor, Anthropic, to buzzy early-stage companies similar Safe Superintelligence.

There is besides a increasing precedent for high-level ex-OpenAI folks to go seed-stage investors. About a twelvemonth ago, Peter Deng, OpenAI’s erstwhile caput of user products joined Felicis. He’s been crushing it ever since, and intelligibly having fun, getting successful connected large deals for blistery startups similar LMArena and Periodic Labs.  

“I really had a telephone with Peter a fewer months ago, and helium helped maine marque the decision,” Rosenthal said of her prime to go an investor. 

But Rosenthal whitethorn person a concealed limb to triumph deals. She besides has heavy contacts among AI endeavor users – the benignant of buyers and beta testers these aboriginal AI startups need. 

Enterprises inactive don’t recognize however overmuch AI tin bash for them. “There’s a truly ample spread that I americium precise optimistic tin beryllium filled. It leaves a immense greenish tract for applications and companies.”