AI Accelerates Incubation Velocity

I came to Primary five years ago to build an incubation engine. VC friends thought it was a fool's errand because, “incubation doesn’t work.” Then Snowflake went public, and I began exploring the overlooked history of successful incubations: Pure Storage, Affirm, Moderna, MongoDB, Palo Alto Networks, and Tinder, to name a few. I realized that incubation works, but it’s hard.

It takes commitment. Then came ChatGPT, and we realized that rigorous, steady, incubation powered by AI would allow us to do something new. We started with simple experiments like writing outbound campaigns with Jasper and experimented with early versions of Clay for automated email flows. Today, we have a proprietary AI-powered Incubation Operating System (LabsOS). With each new model, we’ve gotten more sophisticated and it’s getting exciting. Today, we’re celebrating 20 incubations powered by Primary LabsOS.

Dario Amodei, CEO of Anthropic, likes to talk about PhD-level AI, and in a sense, we’re trying to create a PhD-level incubation thought partner for our check-writers and Operators-in-Residence (OIRs), who together drive on specific incubations.

What does an AI incubation thought partner look like?

LabsOS is a focused suite of agents, tools, data, and Primary playbooks that helps our founders manage the incubation process end to end using a simple chat interface. This includes:

• Thesis Graph: Based on your area of interest as an investor or high level thesis, create a comprehensive knowledge graph of relevant academic papers, commercial information, and patents. Interact with this using LLMs to uncover trends and ideas. Guide a team of agents to integrate the information to generate incubation ideas.
• Radar: Get the recent applicable information across news, earning calls, startup funding, social announcements, and talent movement in your inbox weekly.
• Conversation Agent: Primary’s proprietary system that helps identify the right people, manage outreach, and capture insights from conversations end to end.
• Recruiting Engine: Feed a job packet and get a vetted candidate pipeline in under two minutes, especially helpful for niche and technical roles.
• Co-pilot: A chat interface which has access to all data relevant to your incubation including the outputs of the agents above, like internal and external calls and documents.

How is it helping us concretely?

Let’s use an example. We are currently working on an AI infrastructure incubation and putting LabsOS to work.

The Thesis Graph agent ingested our semiconductor thesis and built a structured knowledge graph, comprising 7000 academic papers, hundreds of patents, and commercial reports. The agent then organized the data across our areas of interest—data center, rack, server, and chip layers, filtering out irrelevant information. What once took months of manual research is now accomplished in hours, creating a corpus that’s fully searchable in natural language and more technically detailed than off-the-shelf LLMs. Here’s an example of a technical output:

Using this corpus, the agent can uncover pre-commercial research trends and map emerging hubs of technical talent across universities and companies, helping the team build relationships with the leaders behind them.

Our diligence tools also save time and elevate work. We are able to systematically find extremely niche experts using simple natural language. For example, questions like “find me 50 integration-focused architect leaders at leading semiconductor companies?” or “who recently left Tesla Dojo’s team?”

Finding quality operators and founders is crucial for a successful incubation. Our recruiting agent understands Primary’s recruiting preferences and builds an initial pipeline in under two minutes, down from several hours. It also uncovers talent we previously might not have been able to find. For example, most recently, we went from a job description to 100+ candidates for a CTO position, with a ~60% pass‑to‑screen rate. This also works especially well with niche talent requirements such as our recent search for ex‑Supermicro enterprise sellers (matched to our ICPs) who might be looking for a new role.

The bottom line is, LabsOS allows Primary to support building new companies at an unprecedented speed and scale. We plan to heavily invest in this system over the next year and scale it out to other incubations and investments at Primary.

Why does this work? Founder-driven incubation.

In many ways, Labs has one goal: recruit incredible CEO-founders. We have learned that accelerated timelines, rich insights, rigorously researched, supported by early design partners and technical de-risking, attracts incredible CEOs. LabsOS is entirely built to support the founder and help them achieve incredible outcomes. And we take it beyond the idea. When you incubate with Primary, the founder is the founder, full stop. In terms of equity and, just as importantly, de facto and de jure control, the founder is just like a founder in our portfolio.

And it works. On average our incubated businesses raise more than $10 million by seed and over $60 million by Series A, with a 80% seed to Series A graduation rate, far outpacing industry norms.

The First 20—and What’s Next

With companies like Tabs (AI powered revenue automation), which just announced their Series B, and Cake (AI development platform) already out in the world, the Labs team is now preparing to launch its next wave of ventures—all built with the same repeatable, AI-accelerated process.

Our Primary Labs team encompasses a Director of Operations (Bridget Flaherty), Designer-in-Residence (Savannah Wolf), Executive Recruiter (Barbara Aprilakis) and, of course, an AI-builder (Tanmaye Bhatia).

We have new additions to the team who will help fuel this work, too. Jason Lango is joining as our first Silicon Valley-based Incubation Partner focused on AI Infrastructure incubations. Jason has incubated two businesses with Sutter Hill Ventures, working closely with the legendary Mike Speiser. He most recently founded Luminary Cloud, which is a Physics AI platform for rapid design iteration, design exploration and optimization. Luminary just announced their Series B led by N47 with participation from N Ventures (Nvidia’s venture capital arm) and Sutter Hill Ventures.

Marcela Sapone has also joined us as a Venture Partner focused on Vertical AI incubations. Marcela started her career at McKinsey and Bridgewater Associates and is the acting Chair and Co-Founder of Alfred, a residential asset management platform, which has raised $225M. She’s already incubated a business through Primary Labs that aims to become the control layer for the physical economy—more on that coming soon.

Primary Labs would not be what it is without our Operators-in-Residence. We are proud to partner with outstanding OIRs, past and present, who help us create category defining businesses. Thank you to Austin Simon, Co-Founder of Relay Commerce, Iacopo Santini, Co-Founder of Naya Homes, Rebecca Schwartz, Co-Founder of Tabs, Misha Herscu, Co-Founder & CEO of Cake AI, Andrew Blum, Co-Founder of HumanX, Ben Waters, Co-Founder of LightTable, Wyatt Bramhall, Co-Founder in Stealth, Samir Koppolu, Co-Founder in Stealth, Ajay Sridhar, Current OIR, Doug Steinberg, Current OIR and Ross Pantone, Current OIR.

If you’re interested in building LabsOS, joining as an OIR, becoming a founder of an incubation, or are simply fascinated by how startups will be built in an AI enabled world, please email labs@primary.vc.