AI agents do the heavy lifting, people should own the judgement

Head of the Agentic and Generative AI for Insurance conference in New York on November 18-19 2026, Intelligent Insurer spoke to keynote speaker Josh Valdez, Chief Product & Technology Officer, CCC Intelligent Solutions.

1. What do you mean by an "agentic claims ecosystem," and why is it becoming important now?

From an outsider’s perspective, an insurance claim looks like a straight line from first notice of loss to settlement, but anyone who has worked on one knows that's not how it goes. A single auto claim, for example, involves hundreds of decisions spread across carriers, repairers, parts suppliers, medical providers, and the policyholder, with handoffs at every step. Today, almost every point in that network is a person. In an agentic claims ecosystem, AI agents join the network as participants. They understand the context, act within clearly defined authority, and pass work to the right person or system at the right moment.

This approach has the potential to support our industry at a time when two significant pressures are colliding: 

1. The work keeps getting more complex -- vehicles with ADAS and new materials, higher repair and injury severity, and more data on every claim. 

2. Experienced adjusters and estimators are retiring faster than anyone can replace them. The industry can't discount its way back to affordability. It must change the economics of the work itself, and the technology can finally act instead of just advising.

2. How does agentic AI differ from the generative AI tools?

Generative AI mostly produces things. It summarises a demand package, drafts a letter, or answers a question. That's useful, but a person still has to take the next step. Agentic AI goes from insight to action. An agent understands the goal, works out the steps, calls on other tools and agents, and advances work within insurer-specified guardrails.

I like to ask a room of people “How many of you have told an agent to do something on their behalf today?” Very few hands go up, and that’s the gap. Take a collision repair estimator, for example. An agent can label photos for each insurer’s specific requirements, pull the plate number and vehicle build data, merge the assignment, and call a parts-sourcing agent for recommendations. The estimator reviews and decides. Because agents act, the bar for accuracy goes up. In this industry, a fast, but wrong answer is the most expensive answer there is.

3. Where do you see the biggest opportunities for AI agents to improve the claims journey?

I’d push back a bit on the idea that the opportunity is “everywhere”. Plenty promise to handle everything from FNOL to resolution. What insurers actually want is specific: name the problem you solve and the outcome that changes.

I’d start where errors are most costly downstream. The estimate is a good example -- its quality affects supplements, parts, cycle time and customer satisfaction. An agent that flags missing operations and checks repair procedures before an estimate is submitted removes a lot of back-and-forth. At FNOL, agents can run a live conversation with the claimant, verify identity and coverage, start predictions on the extent of the damage, and route the claim to the right repairer under the carrier’s rules.

In the casualty space, agents can turn a 300-page demand package into facts with citations back to the source. In fraud, it can catch signals a person could miss. Across all of these, the value comes from giving adjusters back the time they now spend gathering and organising information so they can use it on decisions and people.

4. What will it take for insurers to move from AI proofs of concept to production at scale?

Most pilots stall even though the model works, because the model isn’t connected to the data, context, governance and workflow where the decision actually gets made.

Getting to scale takes a few things: first, start with a use case rooted in real operational pain instead of trying to boil the ocean. Second, measure outcomes, not outputs: cycle time, accuracy and cost, not how many summaries were generated. Third, build on a foundation of existing data, so each new agent doesn’t start from zero. Then, put that agent into the systems and workflows people already use. Finally, get something into production and learn from it. One agent working on real claims will teach you more than ten pilots.

5. How should insurers balance AI-driven automation with human decision-making, particularly for complex claims?

I don’t see it as a trade-off. It comes down to defining authority. Agents should take the repetitive coordination work that buries people today. People should own judgment, exceptions and accountability. Our view is that the future of this industry depends on scaling the expertise of people, not replacing it.

Complex claims make this especially clear. With auto physical damage, casualty, and on-the-job disability claims, the agent’s job is to bring up every relevant fact with a link to the source document and to flag inconsistencies. The adjuster then makes the call with a complete picture. I’d ask the same questions of any AI agent: What can it see? What can it do? What limits it? What evidence does it give? Where does a person step in? If you can’t answer those clearly, the agent isn’t ready for a complex claim.

6. What are the biggest challenges around data, governance, security and trust that insurers need to address?

Data comes first. AI is only as good as the operational data it’s built on. A general-purpose model doesn’t understand repair procedures, parts or injury treatment. Model providers bring the AI, carriers, repairers, OEMs and other P&C ecosystem providers bring a deep knowledge of their own respective operations, and the gap is the industry-wide context that links the two.

On governance and security, agents that act across systems and organisations need clear permissions, defined authority, audit trails and explainable decisions. In a regulated industry where decisions carry real consequences. Trust is the hardest piece. It’s earned through accuracy and consistency over time. Anyone can buy an AI model -- trust that’s built across an entire interconnected network takes decades.

7. How could agentic AI change the role of claims professionals and the way claims organisations operate?

Claims professionals will begin directing the work of people and agents, check that outcomes are right, resolve exceptions, and improve the process over time.

The real advantage will come from putting more expertise into every working hour, not from adding more working hours. In practice, that means your most experienced people can cover more because routine work no longer fills their day. For the organisation, the operating model has to change along with the technology. Data, workflows, governance, and people need to move together. Putting an agent on top of a broken process just produces wrong answers faster.

8. For insurers exploring generative or agentic AI today, what should they be doing to turn experimentation into business impact?

Be specific and be honest about where the pain is, choose the decision that matters most to your economics, and define the outcome metric before you build anything. Bring in the people who know the process end-to-end, not just the AI team -- that includes operations, process owners, IT, and transformation leaders.

Then choose your path with purpose. Some insurers will build their own agents, some will work with partners, and some will deploy ready-made agents. Most will do a little bit of each. Whatever the path, every agent should rest on a trusted foundation of industry data, connected workflows, and governance. 

9. Is there anything else you would like to highlight ahead of your keynote at the Agentic and Generative AI for Insurance conference?

My session is called “From Models to Movement”. The point is simple -- the industry already has plenty of powerful models. What it lacks is the connective tissue that turns them into coordinated action across the claims ecosystem. At the Agentic & Generative AI for Insurance conference, I’ll be talking about the foundation that makes AI work at scale, how to build trust through clear authority and oversight, and where agents can make a real difference for consumer affordability, customer experience, and operational performance. The most useful conversations I have about agentic AI start with someone's actual problem, not with the technology, so I’d encourage anyone in New York to bring those tough questions!

Did you get value from this story?  Sign up to our free daily newsletters and get stories like this sent straight to your inbox.