The business of insuring an explosives manufacturer is, understandably, not fast. A commercial insurance broker navigating that placement juggles carrier portals, ACORD forms, underwriter emails, and rounds of negotiation—a process that can stretch across two weeks and as many as 50 discrete steps. Vincent Chen, a former Google machine learning engineer, watched this unfold and saw something else: a workflow ripe for automation.
His answer is Panta, a four-person startup that doesn't sell software to brokers. It is the broker—one that claims to run nearly its entire back office on AI agents operating from a rack of Mac Minis.
The company emerged publicly in late February from Y Combinator's Winter 2026 batch, positioning itself not as a productivity tool for the insurance industry but as a licensed brokerage where autonomous agents log into carrier systems, fill out forms, email underwriters, negotiate terms, issue certificates, and process renewals. Humans remain in the loop at high-stakes decision points, Chen says, and the system keeps full audit logs. But the pitch is unambiguous: what takes traditional brokers two weeks, Panta believes it can do in days.
Whether that holds up under scrutiny—regulatory, technical, or operational—remains to be seen. The company has been live for roughly three months.
Targeting Risks Nobody Else Wants
Panta is going after the excess and surplus lines market, a corner of commercial insurance reserved for risks that standard carriers won't touch. (The company pegs the market at roughly $125 billion, though that figure hasn't been independently verified.) Think hazmat trucking fleets, cannabis growers, steel mills. These are placements that require specialized knowledge and, often, a broker willing to work the phones and portals until something sticks.
Chen and his cofounder, Frank Wang—a senior full-stack engineer from Apple and a top contributor to the Rust programming language—both hold commercial broker licenses themselves. Panta operates under a California surplus lines broker license (number 4512382, for those keeping score) and claims to have secured licensing across all 50 states, though independent verification of that broader footprint wasn't immediately available at launch.
The technical approach sits somewhere between pragmatic and ambitious. Rather than building workflow software for existing agencies, Chen and Wang constructed what Chen describes as an "OpenClaw-style" agent—software that doesn't just automate tasks but actually navigates carrier websites, manages file systems, and handles the coordination chain that, by Chen's estimate, eats up about 80 percent of a broker's time.
"We're not making brokers more efficient," Chen told a small group at the YC launch. "We're replacing the part of the job that's basically being a human API."
The company's marketing materials cite a case study that hints at the model's potential: bindable quotes delivered in three days for a high-risk armed security firm protecting a 100,000-person stadium—a placement that generated 60 emails. Other anecdotal wins mentioned include explosive oil and gas labs and hazardous gas facilities. One line on Panta's website stands out for sheer audacity: the company says it wants to "insure a rocket" this year.
A Suddenly Crowded Space
Panta's launch coincides with what looks like a mini-boom in AI-native insurance ventures. Late February saw Harper close $46.8 million in combined seed and Series A funding. Earlier that month, Gyde launched with backing from Lightspeed, branding itself as the first "AI-native brokerage platform." WithCoverage, which raised $42 million in February, is applying AI to audit workflows within its own brokerage model.
What sets Panta apart—at least in its pitch—is that it's fully vertical. It doesn't sell tools; it underwrites the service promise itself. The company claims access to more than 100 A-rated carriers through what it calls a national network, though it hasn't disclosed specific carrier partnerships. Marketing claims include a 99 percent placement rate and 24/7 support, assertions that lack third-party validation but suggest how the founders see their competitive edge: speed and specialization in a market where both increasingly matter.
Chen's background adds credibility to the technical story. Before Panta, he led research on recommendation foundation models at Google's Vertex AI and contributed to NotebookLM. Wang built generative AI chatbots and Rust infrastructure that handled over a million real-time messages monthly at Apple. Both earned their broker licenses before founding the company last year—a detail that signals seriousness, or at least awareness that insurance is as much about trust and regulation as it is about technology.
What They're Selling, and to Whom
Panta focuses on what it describes as industries "building the world": construction, transportation, manufacturing, agriculture. There's also a dedicated page for cannabis coverage, a sector that remains challenging to insure in many states. The product menu spans general liability, commercial auto, workers' comp, property, professional liability, excess and umbrella, cyber, D&O, and product liability.
Operationally, the company functions as a California surplus lines broker, a designation that allows it to place coverage with non-admitted carriers when standard markets pass. California surplus lines business carries a 3 percent tax and a 0.18 percent stamping fee—administrative details Panta discloses in its FAQ, perhaps to head off questions about cost structure.
The team remains lean. Y Combinator's directory lists four people; LinkedIn suggests somewhere between two and ten employees, a range that probably reflects how startups count contractors and advisors. No outside funding beyond Y Combinator's standard investment has been announced. Demo Day is set for March 24.
The Open Questions
Can an agent-driven model scale across 50 regulatory jurisdictions? Will carrier relationships hold when the entity on the other end of the email isn't, strictly speaking, human? What happens when a truly complex placement—one involving unique contractual nuances or judgment calls—hits the system?
Three months of production data isn't enough to answer those questions definitively. And the insurance industry, for all its inefficiencies, tends to reward experience and relationships, not just speed.
Still, the underlying thesis is hard to dismiss. If brokers really do spend the majority of their hours as what Chen calls a "human API"—shuttling data between disconnected systems, chasing underwriters, filling out forms—then automating that layer could unlock real margin and service-level improvements. In a market where businesses routinely wait weeks for coverage on essential operations, cutting that timeline to days wouldn't just be a product tweak. It would reset expectations.
Whether Panta can deliver on that promise at scale, and whether the insurance establishment will let it, is the story still being written.
