I spent nearly two decades inside one of the largest commercial insurers in the country. I watched technology reshape personal lines underwriting from the inside and for much of that time, the conventional wisdom held that commercial and specialty lines were different. Too complex. Too heterogeneous. Too dependent on judgment that couldn't be encoded in a rule set. That conventional wisdom wasn't wrong. It was just early.
Today, when I look across property and casualty carriers, I see a significant gap between the potential of AI in underwriting and actual adoption. The headlines suggest a revolution already underway. The reality inside most carriers is more measured with a lot of pilots, a lot of enthusiasm, and a culture that has spent decades developing something it rightly believes is its competitive edge. That culture is worth taking seriously. But here's the thing about underwriting knowledge: if it can be learned, it can be trained. And if it can be trained in a human being, technology should ultimately be able to replicate it. The question was never whether it was going to happen, it was when.
The reason deterministic rules-based systems could transform personal lines but couldn't crack commercial underwriting comes down to one word: heterogeneity.
Think about two houses sitting next to each other. They hold families. They serve substantially similar purposes. The risk profile has a structure you can work with. Now think about two businesses sitting next to each other. One might be a tax preparation office. The other might be a small manufacturer doing custom parts fabrication. Same street, same block — entirely different risk universes. That heterogeneity is why you couldn't write a rule set comprehensive enough to replace a commercial underwriter. The edge cases weren't edge cases. They were the job.
What changed the equation is compute. Deterministic rule sets were always limited by the data you could store and the compute you could bring to the problem. Cloud computing removed the storage ceiling. AI removed the reasoning ceiling. For the first time, we can bring genuinely massive compute to heterogeneous problems and have the system reason through them rather than match them to a predefined pattern.
A copilot that surfaces information for an underwriter to act on is useful. An agentic AI system that operates independently across the underwriting workflow is something categorically different.
Underwriting is not one decision, it's a sequence of discrete processes: submission intake, risk identification, data gathering, appetite assessment, pricing, referral routing, documentation. Each step is a candidate for full or partial automation. Agents that don't just assist but act — retrieving third-party data, flagging appetite mismatches, drafting coverage terms — working across systems without waiting for a human to move them along. That's the shift.
We are at a plateau moment in AI development. The successive leaps that gave us each new model materially surpassing the last have stabilized. That stabilization is actually good news for enterprise adoption, it means moving forward with the technology as it exists today is a rational decision. You don't face the same risk that by the time you've implemented something, the next shift has rendered your investment obsolete.
The plateau is the green light. Carriers who treat this moment as a reason to wait are misreading it. This is the window.
For underwriting leaders at commercial carriers, I'd offer this: we've been through this before. What's happening in commercial and specialty lines today is the same transformation personal lines went through twenty or more years ago. Technology caught up with the complexity of the problem. The industry adapted. The role of the underwriter didn't disappear it just evolved.
The question now isn't whether AI will transform commercial underwriting. It will. The question is how we do it in a way that serves everyone involved: the customers who depend on this product, the agents and brokers who distribute it, and the hundreds of thousands of professionals whose expertise built this industry.
Done right, agentic AI doesn't replace that expertise. It amplifies it.
Bill Devine is Co-Founder and Managing Partner of Naitiv Partners, an AI-native ServiceNow consultancy focused on insurance and financial services transformation. He spent nearly two decades as a senior executive at Travelers Insurance and serves as a former Director of ACORD Standards.
.svg){kind=icon}
