I just got back from one of the largest insurance technology gatherings in the United States. Six thousand people. Two days. More AI pitches than I could count.

Here is what I actually took from it.

The term “AI” has stopped meaning anything.

By midday on Day 1, AI had become background noise. Every booth. Every keynote. Every panel. AI-powered this. AI-native that. I watched experienced carrier people—underwriters, claims directors, CIOs who have been in this industry for 20 years—glaze over in real time.

That is a problem the industry created for itself. When everyone claims the same capability, the claim stops traveling. What cut through was specificity: a vendor who could describe exactly which step of a claims workflow they were touching, why that step, and what the adjuster’s experience looked like after. Those conversations were short because they didn’t need to be long.

The rest were noise.

Most of the “innovation” on the floor already exists inside carrier walls.

This is the observation that struck me hardest. Walking the expo, I kept recognizing what was being demonstrated. Claims automation. Underwriting decision support. MGA process tooling. Loss ratio dashboards. All of it framed as the next wave.

We have already built versions of these things inside our clients’ actual systems. What I was seeing on the floor was a cleaner interface on a problem our clients have been living with for three years. A demo that works against sample data is not the same as a solution that works against a Guidewire implementation with 12 years of customization on top of it.

The startups who didn’t understand that distinction were easy to identify. When a carrier subject matter expert started asking real questions, you could see the moment the conversation exceeded what the vendor had prepared for. Some recovered. Most didn’t.

Carriers are not resistant to change.

They are resistant to being burned again.

There is a version of this story where the insurance industry is the slow, conservative actor and the startups are the agents of progress. That story was plausible in 2019. It is not accurate in 2026.

What I observed is an industry that has run hundreds of pilots, spent real money, and in most cases has little production deployment to show for it. Fewer than half of insurance businesses have gotten AI into a single operational function. (Source: “Insurance Industry Spending Billions on AI With Little to Show for It, New Research Finds from Simplifai,” April 2026)

The carriers are not the problem. The pattern of vendors who promise transformation and deliver a proof of concept—that is the problem.

The result is a procurement environment where trust is the actual currency. A carrier who has been through two failed modernization programs is not going to hand their core systems to a vendor with a good deck. They are going to hand them to someone who has done this before, in a real insurance environment, and can prove it.

That shift is already happening. It is not coming.

The human factor is not a soft point. It is the hard constraint.

Insurance has a knowledge problem that AI cannot solve on its own. A generation of experienced professionals—underwriters who know the edge cases, claims people who understand how coverage arguments actually play out, operations leads who remember why a particular workflow was built the way it was—is leaving the industry. The institutional knowledge encoded in platforms like Duck Creek, Sapiens and Majesco was put there by people. When those people retire, the documentation does not cover what they knew.

Every vendor at the conference who could demonstrate genuine insurance domain knowledge—not “we have an insurance vertical” but actual working familiarity with how coverage terms translate to system behavior, how claims reserving logic works, what an MGA’s data environment actually looks like—had a different quality of conversation than the ones who couldn’t.

AI makes that expertise more valuable, not less. The tools get faster. The judgment required to deploy them correctly stays human. That is not a temporary condition.

Insurance is a smaller world than it looks.

That changes everything about how you operate in it.

This is not an insight you find in market research. It is something you only understand after spending time inside the industry.

The global insurance market is enormous. The population of people who actually make technology decisions inside it is compact enough that at any serious industry gathering, you will find mutual contacts within two degrees of almost any conversation. A carrier CIO in New York has worked with someone who knows your contact at a London MGA. The wholesale broker you met the first day turns out to share three former colleagues with a carrier you’ve been working with for two years.

In a community this connected, reputation is not a marketing asset. It is operational infrastructure. A failed implementation does not stay in one client’s files. A successful one does not stay there either. This is why the carriers and MGAs and wholesale brokers who are making the best technology decisions in 2026 are not responding to inbound pitches. They are calling people they trust, who call people they trust, until they find someone who has done the specific thing they need to do.

If you are not already inside that network, a conference booth will not put you there.

What this means.

The AI moment in insurance is real. The window for modernization is real. The urgency is real—legacy systems are actively limiting what carriers can do in a market that is moving faster than their infrastructure was designed to handle.

What is also real is that the industry has seen enough failed promises to know the difference between a vendor who understands the problem and one who has built a product they believe can be applied to it. That distinction is now the first filter in most serious procurement conversations.

The vendors who are winning in 2026 are not the ones with the most visible AI capability. They are the ones who have been inside enough insurance environments to know what they are actually being asked to do.

Featured image: AI-Generated (ChatGPT)