We Built an AI-Native Insurer. Here's Why Incumbents Can Win Too.
Recently, there's been talk from AI-native insurance startups telling incumbents they'll never catch up. The argument goes like this: The barrier isn't technology; it's organizational DNA. Boards resist. Agent networks resist. Incentive structures resist. Even superintelligent AI can't rewrite a captive distribution network or a CEO's risk tolerance.
We built one of those AI-native insurers. We've spent nearly a decade learning where AI actually works in insurance—and where it doesn't. So we'll say what most people in our position won't:
The critics are only half right.
The organizational immune system is real
We've watched it operate from the inside.
AI threatens more than processes. It threatens people, hierarchies, and decades of institutional knowledge that leaders built their careers on. The more powerful the technology gets, the more threatening the disruption feels, and the harder the organization pushes back.
The execution gap is genuine too. Deloitte surveyed 3,200 enterprise leaders this year and found that executives feel strategically ready for AI but not operationally ready. Every insurance business we talk to confirms this. The board said yes. The pilot worked. But not much actually changed. They tripped in the last mile.
If you're reading those blog posts and feeling uneasy, trust your instincts. Standing still is falling behind.
Where the thesis breaks
The "incumbents are dead" argument assumes the only way to win with AI is to have been born with it. That organizational barriers are permanent. That traditional insurance businesses are evolutionary dead ends waiting for the asteroid.
This confuses two different problems.
The first is building AI technology. AI-native startups have a real advantage here. Clean architectures, ML engineers who learned to work alongside actuaries, feedback loops from day one.
The second is having the operational context that makes AI actually work in insurance. Here, traditional businesses have an advantage no startup can replicate.
A startup can build a great claims model. But it doesn't know that your Florida BI team handles litigation differently than your Texas team because of venue-specific judicial considerations. It doesn't know that your underwriting knowledge base says one thing but your senior underwriters do another—and the deviation is actually producing better results. It doesn't know which of your 50 state regulatory constraints are real compliance requirements and which are institutional habits nobody has revisited in a decade.
That operational context—the messy, human, state-by-state reality of how insurance actually works—is the raw material AI needs to generate value. Technology is the engine. Context is the fuel. Insurance businesses have been accumulating this fuel for decades.
The startup pitch is: "We have the engine, and we'll figure out the fuel." The honest answer is that the fuel is harder to build than the engine.
The real question is speed
Can you close the execution gap before it shows up in your results?
The gap closes by connecting AI to the operational reality of how your business actually runs—across claims, underwriting, distribution, and compliance—in ways that compound over time. Every month of operational AI data makes the system smarter. Every feedback loop accelerates the next one. This is an exponential curve, not a linear one. The businesses that start building now aren't just catching up. They're beginning a compounding process that gets harder to replicate with every cycle.
We spent nearly a decade building these feedback loops inside our own company. That experience made one thing clear: the distance between an AI demo that works and an AI system that changes how you operate is almost entirely about understanding the insurance underneath.
What I'm telling insurance executives right now
Your data is an asset that will appreciate with use. Your operational context can be your advantage. The AI-native startups telling you it's over are talking their own book.
Some businesses already know this. The ones investing seriously in operational AI—not pilots, but production systems touching real policyholders—are proving the thesis wrong in real time. We're seeing this from carriers, MGAs, and specialty businesses alike.
But the window is real. AI feedback loops compound. The businesses that start building them in the next 12 to 18 months will pull away from those that don't. You'll see it first in expense ratios, then in loss ratios, and then in competitive position.
The businesses that win won't become AI companies. They'll stay insurance companies that figured out how to make AI compound inside their operations before the window closed.
Kyle Nakatsuji is the founder and CEO of Clearcover and founder of Dearborn Labs. His team has been building AI-powered insurance operations since 2016.
// Key Questions
Can incumbent insurance carriers compete with AI-native startups?
Yes. Incumbent carriers possess operational context that no startup can replicate -- the state-by-state claims handling patterns, underwriting judgment calls, and regulatory nuances built over decades of writing policies. AI-native startups have cleaner technology architectures, but technology is the engine and operational context is the fuel. Carriers that start building AI feedback loops into production operations now can close the gap, because every month of real operational data makes their systems smarter in ways a startup without that book of business cannot match.
What is the AI execution gap in insurance?
The AI execution gap is the distance between a successful AI pilot and a production system that changes how an insurance operation actually runs. Deloitte's 2026 survey of 3,200 enterprise leaders found that executives feel strategically ready for AI but not operationally ready. In insurance, this manifests as demos that impress the board but never touch a real policyholder -- because connecting AI to the messy reality of claims workflows, underwriting guidelines, and 50 different state regulatory environments is harder than building the model itself.
What is an AI feedback loop in insurance operations?
An AI feedback loop is a system where operational outcomes automatically improve the AI models that produced them -- creating a compounding cycle of better decisions over time. In insurance, this means claims data improves the underwriting model, which improves risk selection, which produces better claims data, and so on. These loops are exponential, not linear. A carrier that starts building them 18 months before a competitor doesn't end up 18 months ahead -- it ends up on a fundamentally different trajectory, visible first in expense ratios, then loss ratios, then competitive position.
Why do most insurance AI projects stall after the pilot phase?
Most insurance AI projects stall because they underestimate the operational complexity beneath the technology. A claims triage model works in a demo, but deploying it in production requires understanding that your Florida BI team handles litigation differently than your Texas team, that your senior underwriters deviate from the knowledge base for good reasons, and that some of your state regulatory constraints are real compliance requirements while others are institutional habits nobody has revisited. This operational context -- not the AI model -- is where most projects break down.
What is the organizational immune system in AI adoption?
The organizational immune system is the institutional resistance that activates when AI threatens existing processes, hierarchies, and career-defining expertise. It's more than change resistance -- it's a rational defensive response from people whose roles, authority, and institutional knowledge are directly threatened by automation. The more powerful the technology, the harder the organization pushes back. Overcoming it requires framing AI as a tool that amplifies existing expertise rather than replacing it, and starting with production use cases that make current employees measurably better at their jobs.
How long do insurance carriers have to adopt operational AI?
The practical window is 12 to 18 months. AI feedback loops compound exponentially, so early adopters don't just get a head start -- they get an accelerating advantage that becomes harder to close with each cycle. Carriers investing in production AI systems today (not pilots, but systems touching real policyholders) are already showing results. Those that wait will see the gap appear first in expense ratios as operational efficiency diverges, then in loss ratios as underwriting models improve, and finally in competitive position as the compounding effect becomes visible to distribution partners and regulators.