Insurance is a probabilistic problem. For forty years, the industry solved it with deterministic software, because deterministic software was all that existed. Every form, every rating rule, every "that's just how insurance works" is a compromise with tools that couldn't do the actual job.
The tools changed. We built a company on the assumption that they would.
What a Neo-Insurer Is
MGT is a neo-insurer. The term deserves a plain definition, so here it is: a full-stack insurance company — licensed, regulated, carrying real risk — built from the ground up on agentic AI, run by a deliberately small team, where the machinery of the company is AI that reasons rather than AI that routes. Not a technology layer on top of a carrier. Not a carrier with an innovation lab. The whole thing, rebuilt from the goal backward.
That was a decision, not a discovery. My co-founder Michael Topol and I started down this path before any of it was proven, and most of the smart money said the same thing: come back after you've done it. The regulatory moat that keeps challengers out of insurance is real, and it's exactly why incumbents haven't had to change. They're protected from the outside and locked in from the inside, running on systems so old nobody left remembers why they work. We decided the moat was worth crossing precisely because so few would follow us across it.
Why We're Writing This Down
We decided the moat was worth crossing precisely because so few would follow us across it.
We're now on the other side, operating. And we've decided this moment is worth writing down, not as marketing but as a record. We're living through the biggest shift in how software gets built since the internet, and we happened to build an insurance company right in the middle of it. Over the coming months, the people who did the work are going to tell you what that actually looked like. Not the highlight reel. The real thing, including what broke, what took longer than we said it would, and what still isn't done.
Three ideas run through everything we'll publish.
Built, Not Bought
When the cost of building software collapses, the buy-vs-build debate collapses with it. Nick Thomas, a principal software engineer on our web apps team, shipped our next-gen policy platform at half the budget we planned. He'll explain what "half the budget" even means when AI is writing the code, because the labor math changed under him mid-project. Robert Haddan, our Head of IT for platform, converted COBOL raters that vendors quoted teams and months for, and he'll also tell you why we still happily pay some vendors. And I'll tell you about Aimee, our broker-facing appetite agent, whose users thank her and ask her to start their quotes, most of them not realizing she's an AI, which raises a question we'd rather answer in public than dodge.
Trust Is Engineered, Not Assumed
Speed means nothing to a regulator, a broker, or a policyholder unless you can prove fast isn't reckless. Matt Evanoff, one of our principal engineers, built an evaluation system that measures every decision our AI makes, and his post will include the part that makes the rest believable: what evals still can't catch. It matters because of what Chris Weed is building in agentic QA, where a bug goes from found to fixed in production in minutes, and "we'll fix it next sprint" stops existing as a concept. That speed is only sane because the verification runs just as fast.
Leverage per Person
The bet underneath everything: one person, working with the right agents, does the work of many. Niall Fitzgerald, our Head of Insurance, isn't an engineer and doesn't write tickets anymore. He builds. Our underwriters will tell you what it feels like when the AI stops being a tool and becomes the colleague at the next desk. And I'll share the moment I'm still not over: the first time Athena, our AI underwriting agent, made an end-to-end decision with no human input, and got it right.
There's a difference between adapting to a shift and being built for one.
This is the part I want you to take from this post: none of it was luck. We designed a company for a world where AI reasons, and then that world arrived. Small team, no legacy, every system built from the goal backward. That's not a constraint we accepted, it's the advantage we chose. There's a difference between adapting to a shift and being built for one, and that difference is the whole reason MGT exists.
We're going to get there first. Here's how.
