The feedback loop took months to complete, but when it did, Vincent Chen and Frank Wang realized they'd been solving the wrong problem entirely.
Throughout much of last year, the two engineers made their pitch to insurance brokers: AI tools that could streamline portions of the placement workflow. Interesting, the brokers would say. Clever, even. But then came the real talk. The problem wasn't this feature or that integration. It was the whole edifice—the logging into carrier portals at 11 p.m., the endless application forms, the email chains with underwriters that stretched into the void, the phone calls that went nowhere. One broker after another described drowning in orchestration.
So Chen and Wang, friends since high school and by then licensed brokers themselves, made what might seem like a peculiar pivot: they stopped building tools for brokers and started building AI agents to become them. The result, a company called Panta, isn't exactly a software product. It's a commercial insurance brokerage where the brokers are mostly code.
Panta quietly went into production late last year and has been operating with software-like efficiency ever since. Whether that's a triumphant reimagining of a centuries-old profession or a brute-force experiment in automation depends, perhaps, on where you sit in the industry.
The Unusual Mechanics
This isn't a chatbot situation. Panta is a fully licensed commercial insurance brokerage—California surplus lines broker, License #4512382—where autonomous AI agents handle the placement process from intake to binding. The agents log into actual carrier portals, the kind human brokers curse under their breath. They complete applications, draft emails to underwriters, negotiate terms in written correspondence, and follow up by phone. Then they do it again. And again. Across thousands of clients simultaneously, if needed.
The company advertises access to over 100 A-rated carriers and claims a 99% placement rate. Its targets are the gnarly risks: construction firms, transportation fleets with questionable safety records, agricultural operations, manufacturers handling materials that make underwriters nervous. The coverage menu spans general liability, commercial auto, umbrella policies, workers' comp, property, cyber, D&O, product liability. Panta says it operates in all 50 states, though verifying that through state insurance department records would require, well, considerably more time than this article allows.
One anecdote from Y Combinator illustrates what speed looks like in this model: an explosives manufacturer, normally a weeks-long odyssey of manual negotiations, was covered in three days after Panta's agents processed 70 pages of forms and fired off more than 100 emails. On LinkedIn, Chen mentioned the agents securing insurance for an explosives oil and gas lab and set a goal for 2026 to "insure a rocket." The rocket is aspirational. The explosives manufacturer, apparently, is not.
The Builders Behind the Agents
Chen comes from Google, where he was an ML engineer working on foundation models for recommendations at Vertex AI. His fingerprints are on products like NotebookLM and Google Photos—the kind of consumer-facing AI that feels effortless because the plumbing underneath is anything but. Wang, who sometimes goes by Jiangda Wang professionally, spent years as a senior full-stack engineer at Apple and ranks as a top contributor in the Rust and Gleam open-source communities. These are people who understand infrastructure.
But here's the wrinkle that might matter most: before founding Panta, they became licensed brokers themselves. Not as a credential-collecting exercise—they actually placed hundreds of policies by hand. Hazmat trucking. Steel mills. Hazardous gas facilities. The unglamorous, teeth-gritting work of commercial insurance. That experience, Chen has said, informed how the agents were architected. During a demo for the Y Combinator Winter 2026 batch, he described the system as "20x" faster than a human broker's submission workflow. Whether that holds across all risk types remains an open question.
The company, according to YC's directory, is a three-person operation based in San Francisco. CB Insights lists a $500,000 convertible note from Y Combinator, though no separate priced round has been publicly announced. Small team, narrow focus, ambitious automation thesis.
The Market They're Wading Into
Panta is entering a commercial insurance market that surpassed $500 billion in direct premiums written in 2024, part of a U.S. property and casualty industry that topped $1 trillion for the first time. Despite decades of talk about digital transformation, independent agents still place 87.2% of commercial lines premiums, according to recent industry data. The distribution model remains stubbornly broker-centric, which means automation that can genuinely collapse the labor intensity of placement might find a receptive audience. Or fierce resistance, depending on who's being automated.
The competitive landscape has gotten crowded quickly. Harper, a YC Winter 2025 company positioning itself as a near-autonomous commercial brokerage, announced a combined $46.8 million seed and Series A earlier this year and reports serving over 5,000 customers. FurtherAI raised $25 million last fall for an AI workspace aimed at insurers and brokers. Anzen secured $16 million around the same time for submission and quoting automation. In December, Willis Towers Watson announced plans to acquire Newfront for up to $1.3 billion, citing the target's "agentic AI" placement capabilities as a key driver of the deal.
A pattern emerges: large brokers are acquiring tech-native firms for their AI infrastructure, while startups are building AI-first models from the ground up, no legacy systems to unwind. Panta clearly belongs to the latter category. Its thesis is that replacing the human back office entirely—not augmenting it, replacing it—is the only path to software-like margins in what has historically been a labor-intensive services business.
What This Actually Means
The bet Panta is making is this: the value in commercial insurance brokerage isn't primarily human expertise or carrier relationships, though those matter. It's the orchestration of tedious, repeatable work that happens to require navigating dozens of different systems, none of which talk to each other. If autonomous agents can truly handle that orchestration across thousands of placements—and do it accurately—the economics shift dramatically. Labor stops scaling linearly. The company claims its agents can automate 95% of backend work, leaving human brokers to handle client relationships and the genuinely complex negotiations that algorithms can't yet navigate.
Whether that vision holds up in the wild depends on things the company hasn't yet publicly disclosed. Error rates, for one. Claim outcomes when things go sideways. Client retention figures, which tend to reveal whether businesses feel well-served or just efficiently processed. And perhaps most critically, how the agents handle edge cases that fall outside their training data—because in specialty insurance, edge cases aren't edge cases, they're Tuesday.
The 99% placement rate is a marketing claim, not an independently verified metric. The assertion that the company operates licensed in all 50 states hasn't been confirmed through state insurance department filings, which are public but scattered and time-consuming to compile. These aren't necessarily red flags—startups often round up when describing their capabilities—but they're worth noting.
Still, some signals feel legitimate. Placing an explosives manufacturer in three days is genuinely fast for a risk that typically grinds through weeks of manual underwriter negotiations and back-channel conversations. The company's focus on surplus lines and specialty risks—markets where standardization is low and human judgment is supposedly irreplaceable—is either wildly ambitious or strategically calculated. Possibly both.
Panta's contact information is straightforward: [email protected], a phone number listed on its coverage page, a San Francisco address that checks out. The website includes a live quote funnel, which means they're taking actual business, not just beta testing in stealth mode. Chen's LinkedIn activity suggests the founders are iterating quickly, treating the product as a proving ground for what agentic AI can do in industries where regulation and complexity have historically kept software at bay.
The insurance industry has resisted disruption for longer than most. Panta is betting that resistance wasn't about the impossibility of automation—it was about waiting for the right kind of automation. Whether autonomous agents are that solution, or just the latest in a long line of promising technologies that discover insurance is harder than it looks, will likely become clear in the next 12 to 18 months. The rocket launch is aspirational. The explosives manufacturer, though, suggests something might actually be working.
