The three-hour walk that built a unicorn
There is a moment - usually unremarkable from the outside - where the trajectory of a career bends irreversibly. For Matan Grinberg, it was a walk through San Francisco with Shaun Maguire, a Sequoia Capital partner who also happens to hold a physics PhD. Grinberg had cold-emailed him, still enrolled at UC Berkeley, working through theorems about black hole singularities and quantum causal models. He wanted career advice. What he got was a complete reorientation.
Three hours of walking, and Maguire laid out the options: join Twitter, join a Sequoia portfolio company, or start something. Grinberg said he might want to build. Maguire told him to drop out and send a screenshot as proof. He did.
I realized that I had spent basically the last decade obsessed with physics, and I was only doing it because it was hard - not because I actually loved it.
- Matan GrinbergThe candor in that admission is worth sitting with. A Princeton BA, a Cambridge mathematics master's degree, and most of a UC Berkeley theoretical physics PhD - all built on a foundation of intellectual stubbornness rather than genuine love. It takes a particular kind of self-awareness to name that distinction while mid-stream, and a different kind of courage to act on it.
72 Hours to a Pitch Deck
Grinberg found his co-founder the way most consequential partnerships seem to happen - at a San Francisco hackathon, in a city that runs on the possibility of accidental collisions between ambitious people. Eno Reyes, Princeton class of 2021, former ML engineer at Hugging Face and Microsoft, showed up in the same room. Grinberg described the connection as "intellectual love at first sight."
Within 72 hours of their first coffee meeting, they had a demo. Not a slide deck. Not a concept. A working demo, ready to show Sequoia. Maguire looked at it, set up a meeting for the next morning, and the pair walked in and pitched. That meeting produced the seed round. Factory was alive.
The founding year was April 2023. The Series C, at a $1.5 billion valuation led by Khosla Ventures, closed in April 2026. Three years. From drop-out screenshot to billion-dollar company.
Not a coding copilot. A software team.
The AI coding assistant market filled up fast. GitHub Copilot, Cursor, Codeium - tools that sit in your IDE and autocomplete lines. Factory took a different bet from the start: autonomous agents that handle entire workflows, not just the next line of code.
Factory calls them Droids. Each one tackles a specific category of software work - writing code, reviewing pull requests, running tests, diagnosing incidents, generating documentation, executing migrations. They operate across IDEs (VS Code, JetBrains, Vim), web browsers, CLIs, Slack, Teams, and project management systems. The tagline captures the intent cleanly: "The only software development agents that work everywhere you do."
The differentiation that Grinberg leans on hardest is vendor agnosticism. Factory's agents can switch between foundation models - Anthropic's Claude, OpenAI's reasoning models, DeepSeek - depending on the task. That flexibility matters in an enterprise context where model selection is a governance question, not just a technical one.
Where the Droids work
The company's enterprise roster reads like a who's-who of organizations for whom software reliability is not optional.
The breadth is intentional. Financial services, chip design, creative software, cybersecurity, pharmaceuticals - these are sectors where legacy codebases run into the tens of millions of lines and migration projects take years. That is exactly where Factory's multi-repo, context-aware approach has the most leverage.
"Pave the roads first"
Grinberg is the kind of founder who gives interviews that contain a useful mental model. His conversation with McKinsey produced one worth keeping: the Ferrari and dirt roads problem.
The Ferrari on Dirt Roads
Giving enterprises a powerful AI agent without fixing their underlying infrastructure is like handing someone a Ferrari to drive on dirt roads. You need documentation, tests, CI/CD pipelines, and observability before the agent can do anything useful. Organizations that skip the infrastructure step find their AI agents stalling in the mud.
The analogy reframes the AI adoption conversation in a way that's useful for executives: the question isn't whether your team is ready for AI, it's whether your codebase is ready for AI. Those are different problems with different solutions.
He also makes a point about organizational clarity. AI agents, he argues, force a kind of accountability that vague human ownership structures can avoid. When a Droid needs to execute a task, it needs to know who owns what. That pressure - properly harnessed - can be a forcing function for better documentation and cleaner architecture, even before any code gets written autonomously.
From seed to unicorn
Factory Funding Rounds
The Series C, closed in April 2026, brought in Khosla Ventures as lead with Keith Rabois taking a board seat. The list of returning and new investors tells the story of how the AI infrastructure narrative has expanded: Sequoia anchored the early rounds, then Blackstone, Insight Partners, NEA, and 20VC joined the table as the category matured from experiment to enterprise infrastructure.
Series C Investor Syndicate
What string theory teaches you about debugging
Grinberg published actual research papers before he built a company. The topics - black hole singularities, quantum causal models - are the kind of problems where the error state is the universe collapsing. That background shapes how he thinks about systems.
Theoretical physics trains you to hold impossibly complex models in your head and locate the single constraint that determines the entire system's behavior. Software architecture has similar structure: vast, interconnected, with leverage points that are non-obvious until someone with the right mental model points at them. The transition from "how does spacetime work" to "how do we make agents reliable in production" is less of a pivot than it appears from the outside.
His co-founder Sequoia partner, Shaun Maguire, also holds a physics PhD. When they went for that three-hour walk, they were speaking a shared language - not just about career theory, but about how to reason under uncertainty. That shared framework probably mattered more than the specific advice.
The policy nobody asked for (but needed)
In early 2026, Factory published a Safe Autonomy Readiness Policy. This was not a regulatory requirement. Nobody was pressuring them. Grinberg published it because he believes the question of how to deploy increasingly capable autonomous agents safely is one the industry needs frameworks for - and he'd rather be a builder of those frameworks than a bystander.
As AI becomes more capable and agentic, it becomes all the more imperative to identify, evaluate, mitigate, and monitor the associated risks.
- Matan GrinbergThe policy is structured around four verbs: identify, evaluate, mitigate, monitor. It covers threat modeling, capability audits, red teaming, stress testing, sandboxing, and human approval gates. It addresses exfiltration risks and the specific problem of agents causing unintended side effects in live systems.
Publishing a safety policy when you're building a product that automates software at scale is not a PR move - it's a statement about what kind of company you're building. Grinberg also makes a point that might seem counterintuitive coming from someone selling automation: AI tools will increase demand for software engineers. The argument is straightforward. Automation raises the ceiling of what's buildable. Every problem that's now too complex to tackle becomes approachable when the baseline productivity of each engineer multiplies. More buildable problems means more engineers building them.
The details that don't make the press release
- Has a personal GitHub project called "breatheHRV" - a breath work helper to improve heart rate variability using Apple Watch. Physics PhD to AI CEO, but also, apparently, tracking his own biometrics with wearables.
- Met co-founder Eno Reyes at a San Francisco hackathon. Described it as "intellectual love at first sight." Their first coffee meeting ended 72 hours later with a working product demo.
- Published research on black hole singularities and quantum causal models before any of this happened. The papers still exist.
- Dropped out of Berkeley after a single three-hour walk. Sent a screenshot to confirm, as instructed. Maguire invested shortly after.
- Both Grinberg and Reyes made the 2025 Forbes 30 Under 30 AI list. Different profiles, same company, same class year.
- Factory's agents are intentionally called "Droids" - the name signals what they are. Not assistants. Not copilots. Things that execute.