Somewhere in San Francisco, a rack of Mac Minis is doing the work of a commercial insurance broker. Not assisting one. Replacing one.
The machines—an unexpectedly humble choice for a company with grand ambitions—are logging into carrier portals, filling out industry-standard ACORD forms, emailing underwriters, and binding policies. Recently, according to the founders, one of these AI agents secured insurance for an explosive oil and gas laboratory. The kind of account most brokers would spend days shopping around, handled autonomously in hours.
This is Panta Insurance Services, a Y Combinator startup that emerged in late February with a premise equal parts obvious and audacious: commercial insurance doesn't need better software. It needs fewer humans.
"Humans are the API," the company's pitch goes. And APIs, it turns out, are slow.
Hard Problems, Fast Answers
The typical commercial insurance placement—especially for what the industry calls E&S, or excess and surplus lines—unfolds like a game of telephone played across time zones. A broker receives an application. Logs into a carrier portal. Fills out forms manually. Sends emails. Waits for quotes. Chases underwriters. Repeats the process across multiple carriers. For complex risks, this can stretch to seven days or more.
Panta claims it can collapse that timeline to roughly four hours.
The company has been in production since December, targeting the gnarliest corner of the commercial insurance market: the risks that standard carriers won't touch. Cannabis operations. High-risk construction contractors. Explosives manufacturers. These aren't accounts you can price with a simple algorithm. They require judgment, specialized underwriting knowledge, and access to non-admitted insurers operating in the surplus lines market.
And yet, Panta insists, an AI agent can handle them.
The company positions itself not as a software vendor but as a full-stack brokerage. It holds a California Surplus Lines Broker license (number 4512382) and says it's licensed across all 50 states. Its product suite covers general liability, commercial auto, property, workers' compensation, excess and umbrella policies, cyber, and directors and officers coverage. Industries served include construction, transportation, manufacturing, and agriculture—the bread-and-butter of E&S business.
According to Panta's technical microsite, its six-stage pipeline averages about eight minutes per stage. The company claims an error rate below 0.1%, a 99% placement rate, and access to more than 100 A-rated carriers. In one week, it says, the operation touched $1.5 million in premium with just 25 calls, placing policies for trucking fleets, demolition contractors, tow operations, and software companies.
Those are marketing claims, not audited figures. But they sketch the outline of something genuinely different: a broker of record handling everything from intake to endorsements, powered by AI that mimics human work at machine speed.
The Human API Problem
The core insight is almost comically simple. Traditional insurtech companies build tools for brokers. Panta built agents that are brokers.
These AI operators log into carrier portals the way a human producer would. They fill out market-specific forms, send emails, make calls to underwriters, handle follow-ups, process renewals. The company describes this as replacing the "human API"—the broker who sits between buyer and carrier, manually translating information across incompatible systems.
For an industry that still faxes documents and relies on PDF email attachments, the technological leap is significant. Perhaps more significant than the founders initially expected.
The technical stack runs on Mac Minis, a choice the company frames with a touch of poetry: "From Mainframe to Mac Mini." Why Mac Minis? The founders haven't elaborated publicly, though the decision suggests a focus on speed and simplicity over enterprise-grade infrastructure. The agents operate with what Panta calls "human approval gates on high-stakes actions," though the specifics of those guardrails remain vague.
Unlikely Brokers
Vincent Chen and Frank Wang don't fit the traditional insurance broker mold.
Chen's background is in AI and machine learning at Google, where he contributed to Vertex AI's foundational recommendation model and NotebookLM Enterprise. He demoed Ask Photos at Google I/O—not the typical résumé line for someone selling general liability policies to construction contractors. Wang came from Apple, where he worked on generative AI chatbots and Rust infrastructure. He claims to be a top-10 contributor to the Rust and Gleam open-source ecosystems.
But both are now licensed commercial insurance brokers, a detail that matters in a heavily regulated industry where credibility depends on understanding underwriting nuance and carrier relationships. You can't just spin up an AI brokerage without navigating state licensing requirements, E&O insurance, fiduciary responsibilities, and the delicate politics of carrier partnerships.
The team is lean—three to four people, according to Y Combinator's directory, operating out of 1301 16th Street in San Francisco. The company raised $500,000 via convertible note, with Y Combinator as an investor. CB Insights lists the round as occurring roughly two months before March 2026, making it recent but not brand-new.
A Market Ripe for Disruption (Again)
The E&S market has been growing steadily. Year-end data from stamping offices across 15 states showed $90.3 billion in premium in 2025, up from $81 billion the prior year. Industry reports have pegged the total U.S. surplus lines market at more than $115 billion in direct premiums as of 2023, though exact figures vary depending on who's counting.
It's a massive, fragmented market. Complexity creates inefficiency, and inefficiency creates openings.
Panta isn't the first to notice. Newfront introduced AI-powered placement tools in 2023 and was acquired by WTW in late 2025. Acrisure appointed a Chief AI Officer in October 2025 and has since reportedly cut accounting positions in favor of automation. Pathpoint offers digital wholesale access aimed at small manufacturers. Fulcrum raised $25 million in January 2026 to automate back-office operations for large brokerages.
The difference, as Panta's founders see it, is that most of these companies are building tools to assist brokers. Panta wants to be the broker—running the full distribution stack with AI agents instead of human producers. It's a riskier bet, one that requires taking on regulatory compliance, carrier relationships, errors and omissions liability, and all the operational overhead of a licensed brokerage. But it could also prove more defensible.
The incumbents are watching. Industry analysts remain cautious, noting that AI deployment in insurance is still early-stage and that fears of disintermediation may be overstated. Brokers, after all, don't just fill out forms. They advise clients, negotiate terms, manage claims, and maintain relationships that span decades. Can an AI agent really do that?
Panta seems to think so. Or at least, it's willing to find out.
The Rocket Problem
The company has stated an audacious goal for 2026: insure a rocket.
It's the kind of headline-grabbing claim that signals both confidence and a willingness to tackle increasingly exotic risks. Rockets are notoriously difficult to insure—complex underwriting, limited carrier appetite, high stakes. If Panta can pull it off, it's a proof point. If it can't, well, it was a marketing line.
For now, the focus is on proving the model works at scale across the less glamorous corners of E&S: construction, transportation, hard-to-place commercial accounts. The stuff that pays the bills.
Whether autonomous agents can truly replace the judgment, hustle, and relationship-building of human brokers remains an open question. The technology is new. The regulatory landscape is uncertain. The competitive response from incumbents is just beginning.
But in a market measured in tens of billions, there's room for experimentation. And Panta is betting that somewhere in that sprawl of inefficiency, there's space for a new kind of broker—one that never sleeps, never forgets to follow up, and can quote 100 carriers before lunch.
Even if it's running on a Mac Mini.
