OpenAI veterans launch $100M VC fund Zero Shot, already flagging AI trends they say are technically unfeasible

The people who built AI’s foundational models are now deciding which companies get to build on top of them. Zero Shot, a new venture capital fund founded by former OpenAI engineers and product leaders, has closed its first funding toward a $100 million target — and its founding partners are already wielding their insider knowledge to declare entire startup categories technically unfeasible. The fund isn’t just another entrant in the AI venture gold rush. It represents a structural shift in who holds power over the AI application layer: the same people who shaped the capabilities of frontier models are now positioned as gatekeepers deciding which bets on those models deserve capital, and which deserve to die.

From builders to allocators

Zero Shot’s five founding partners include three OpenAI veterans — Evan Morikawa (applied engineering leadership), Andrew Mayne (prompt engineering, host of The OpenAI podcast), and Shawn Jain (research and engineering) — alongside Kelly Kovacs, previously a founding partner at 01A, and Brett Rounsaville, formerly of Twitter and Disney. Their advisory board extends the OpenAI footprint further, with the company’s former head of people, former head of communications (who also held that role at Apple), and a former product leader.

The credentials matter less for their prestige than for what they imply: this is a team that has seen the internal roadmaps, understands the architectural constraints, and knows where frontier model capabilities are headed before the market does. That asymmetric information is the fund’s actual product.

First checks written

Zero Shot has invested in three companies. Worktrace AI, founded by former OpenAI product manager Angela Jiang, is developing AI-based management software to help enterprises discover and automate tasks. The startup raised a seed round from investors including former OpenAI CTO Mira Murati and OpenAI’s Fund. The fund also backed Foundry Robotics, which is building AI-enhanced factory robotics and recently closed a $13.5 million seed round led by Khosla Ventures. A third portfolio company remains in stealth.

What they’re avoiding — and why it matters

The more revealing dimension of Zero Shot’s thesis is what the partners won’t fund. Mayne is bearish on most vibe coding platforms, predicting that major model makers will quickly render standalone subscriptions unnecessary. He’s equally skeptical of digital twin startups, having built a reasoning model to test several and concluding that standard LLMs perform comparably.

Morikawa, drawing on his robotics and AI background, is dismissive of startups building embodiment training data from egocentric video, noting significant technical challenges in transferring learning across the embodiment gap.

These aren’t casual opinions. They’re informed by direct knowledge of what frontier models will be able to do in six to eighteen months — knowledge that most founders pitching in these categories simply don’t have. When a former OpenAI engineer says vibe coding platforms will be absorbed by the model providers themselves, that’s not market speculation; it’s a prediction grounded in having seen the product roadmaps that will make it happen. When Morikawa dismisses a particular approach to robotics training data, he’s drawing on architectural understanding of where sim-to-real transfer actually breaks down — the kind of detail that doesn’t appear in published research but shapes billions in R&D allocation inside frontier labs.

The gatekeeper problem

This dynamic has consequences that extend well beyond one fund’s portfolio. Former employees of frontier labs carry asymmetric information about model capabilities, roadmaps, and architectural limitations that outsiders simply don’t have. According to industry observers, predicting where AI models will develop next requires specialized insight, as progress is often non-linear and unpredictable.

Consider the specific mechanism at work. A founder building a code-generation startup goes to raise a seed round. A fund like Zero Shot passes, not because the product is bad today, but because its partners know that GPT-next will ship a native feature that makes the entire category redundant. That signal propagates. Other VCs, seeing that former OpenAI insiders passed, adjust their own assessments. The category cools. Capital flows elsewhere. The insider knowledge doesn’t just inform one investment decision — it shapes which categories of AI companies can exist at all.

The same AI companies generating enormous value through their models are, through their alumni networks, increasingly determining which downstream companies get funded — and which categories get starved of capital. When new AI startups are raising $100 million rounds at extraordinary speed, having former insiders deciding what’s technically viable and what isn’t carries real weight in the market. A vibe coding startup that might have raised comfortably in 2024 now faces a funding environment where the people who built the models it depends on are actively telling LPs that its product category has no durable moat.

Zero Shot’s target funding is modest by current AI fund standards. The signal is less about the capital and more about the source — and about the uncomfortable question it raises: in an industry where the frontier labs already dominate research, talent, and compute, should their alumni also control the capital that decides who gets to build the next layer?

The builders are now the allocators. Whether that makes for better investing or a narrower AI ecosystem is the real bet.

Feature image by Vlada Karpovich on Pexels