Pibit.AI

A submission lands. Somewhere, a machine reads first.

It is Tuesday morning at a mid-market commercial carrier. A 47-page loss run hits an underwriter's inbox along with seven attachments, two ACORD forms, and a broker email that ends with "needs a quote by EOD." Five years ago, this would have started a paper hunt. Today, at the dozens of carriers and MGAs running Pibit.AI's CURE platform, the documents are parsed, the risk is researched, the appetite check is done, and a draft recommendation is sitting in the queue before the underwriter has finished their first coffee.

That is the world Pibit.AI is selling. It is not flashy. It is not consumer. Nobody is writing breathless threads about it on X. But quietly - between the actuarial spreadsheets and the carrier APIs - one of the more interesting agentic-AI companies of the last cycle has been busy putting itself in the wiring of an industry that runs roughly 11% of US GDP.

Insurance is not a data problem. It is a document problem.

The dirty truth of commercial P&C insurance is that the data underwriters need to make decisions already exists - it is simply trapped inside loss runs, SOVs, ACORDs, broker emails, and the occasional handwritten amendment. The industry spent two decades digitizing its policy management systems and almost no time digitizing the messy interface between brokers and carriers. The bottleneck is not analytics. It is intake.

Anyone who has watched an underwriter rekey a five-year loss run from a scanned PDF understands the indignity of the workflow. Anyone who has watched a quote slip past its deadline because the appetite check took three days knows the cost. Pibit.AI's founders looked at this pile and asked an awkward question: why is the bottleneck in the most data-rich industry on earth a series of file types from 1998?

An engineer with an insurance dad.

Akash Agarwal grew up watching his father work the late hours of an insurance agent - the forms, the file folders, the phone calls that always seemed to be about the same paragraph on page 12. Years later, Agarwal would end up at the autonomous-vehicles startup Playment, watching computer vision quietly eat hard problems alive. The contrast was, in his telling, hard to ignore: AI was rewriting how cars saw the road, and his father's industry was still rewriting policy effective dates by hand.

In 2020, with co-founder Karan Bedi, Agarwal started Pibit.AI. The bet was unfashionable at the time - this was the era of consumer AI demos and crypto wallets, not B2B insurance plumbing - but Y Combinator backed the W21 batch and Agarwal got to work selling carriers on a thesis they had heard a hundred times before: we will read your documents better than you can. The difference, this time, was that the tools had finally caught up to the claim.

CURE: five agents that mostly want you to log off early.

Pibit.AI's platform is named, with the audacity only an early-stage founder can muster, CURE. The branding is on the nose. The architecture is more interesting. Rather than one monolithic AI, CURE is five composable modules - each owning a specific stretch of the underwriting lifecycle, each handing off to the next like a well-rehearsed relay.

Numbers the actuaries can argue with.

Insurance is a wonderful industry for AI startups because the buyers do not have to be convinced - they have to be measured. A carrier will not buy a platform because it sounds good in a demo. A carrier will buy a platform because the loss ratio moved. Pibit.AI's published claims are, by insurtech standards, unusually specific.

Customers willing to be named include Kinetic, Shepherd Insurance, HDVI, and Method Insurance Services - a notable mix of MGAs and venture-backed carriers, which is exactly the segment that has the patience for new tooling and the urgency to use it. The $7M Series A from Stellaris Venture Partners in November 2025 is, in this context, a vote less on the technology and more on the customer pipeline behind it.

Make underwriters faster, not extinct.

Pibit.AI's framing is careful here, and the carefulness is itself a tell. The company does not pitch underwriter replacement. It pitches underwriter amplification - the polite term for "we will absorb the parts of your job you hated anyway." This is partly product positioning and partly a clear-eyed read of how insurance buys software: very, very slowly, and only after the humans in the room feel safer with it.

The vision Agarwal describes in interviews is more ambitious than the marketing. An industry in which every underwriter has an agentic co-pilot. An industry in which capacity scales without headcount. An industry in which the loss ratio - that grim, immovable number that defines whether a carrier lives or dies - finally bends because the data behind every decision is richer, fresher, and arrived in time.

If the agents work, the whole industry rewires.

Commercial insurance is one of the largest knowledge-work markets in the economy, and almost none of it has been touched by modern AI. The companies that get this right will not just win underwriting - they will quietly become the operating layer for how risk gets priced. That is a strange and slightly thrilling thing to be building from a Gateway Boulevard office in South San Francisco.

The risks are obvious. Insurance buyers are conservative. The AI hype cycle has trained CFOs to expect overpromises. And the moat in this market is not technology, it is integration depth and trust - both of which take years. But the early indicators are real, the customer logos are real, and Stellaris does not write seven-figure checks for vibes.

Back to the Tuesday morning. The submission has landed. The underwriter, who used to spend the morning fighting their inbox, is reviewing a draft recommendation instead. The work is still hers. The judgment is still hers. The forty-seven pages are someone else's problem now. That is the small, specific, unsexy revolution Pibit.AI is selling - and so far, the people writing the checks seem to be buying it.