Agentic AI Blurring Core System Boundaries—Datos Insights’ Wein

Anthony R. O’Donnell
// May 8, 2026

(Image credit: IIR/Gemini.)

Datos Insights (Boston) is seeing the early signs of a major shift in insurance technology architecture, as agentic AI begins to challenge long-standing assumptions about where underwriting, policy administration and other core-system functions should reside, according to Mitch Wein, Executive Principal, Insurance Strategy and Advisory, Datos Insights.

Wein says the emergence of model context protocol (MCP) as a standard for connecting AI agents to systems and data sources could help insurers move beyond isolated AI use cases toward workflows that span multiple platforms, models and operational domains. That development, he suggests, could blur the traditional boundaries between underwriting workbenches, policy administration systems, claims, billing and other categories that have defined insurance technology for decades.

“The core system vendors are probably in trouble,” Wein comments. “I think they’re going to have to rethink what core is and what it means.”

MCP as Connective Tissue

Anthropic introduced MCP in November 2024 as an open standard for creating secure, two-way connections between AI-powered tools and data sources. In December 2025, Anthropic donated MCP to the Agentic AI Foundation, a directed fund under the Linux Foundation, where it became one of the foundation’s founding projects.

For insurers, Wein says, MCP matters because agentic AI depends on more than a single model or application. Multi-step insurance workflows must move across data sources, systems, models and business rules while preserving context, access controls, lineage and governance. That is particularly important in a regulated industry where carriers handle non-public personal information, health-related data and other sensitive information.

Mitch Wein, Executive Principal, Insurance Strategy and Advisory, Datos Insights.

“You couldn’t string together the workflows,” Wein says of the earlier state of agentic AI. “You couldn’t move stuff from one LLM to another LLM or SLM. It would lose context. The data would get exposed. It wouldn’t retain the rules about the data or the lineage if it was transformed.”

Wein characterizes MCP as a potential connective layer for agentic AI in insurance. Vendors that had been developing proprietary approaches to connecting agentic workflows, he says, are recognizing that insurers will operate across multiple platforms and that agentic workflows will need to communicate across them.

“People aren’t going to be living in just my platform,” Wein says, characterizing the vendor realization. “They’re going to be living in multiple platforms all at the same time.”

The Rise of ‘Skinny PAS’

That shift could be especially important for core systems, whose boundaries have historically been defined with considerable clarity. Datos Insights has long maintained technology maps that distinguish policy administration, underwriting workbench, claims, billing and other insurance system categories. Wein says those distinctions are becoming less rigid as workflow orchestration, data access and AI-enabled decision support begin to operate across the old borders.

“We’re seeing those boxes, those hard boxes become fuzzy,” he says. “The line between the one and the other is kind of merging together.”

One possible result, Wein suggests, is the rise of what Datos calls “skinny PAS.” In that model, the policy administration system becomes more of a record keeper, back-end engine or interface to the ledger, while underwriting, product configuration, distribution interaction and related workflows increasingly occur in workbenches, ecosystem layers or agentic AI orchestrations.

“The policy admin systems are being shrunken down to being more of a record keeper, more of a back-end engine, more of an interface to a ledger,” Wein says. “Then the underwriting and the product stuff and the workflows between the underwriters and the distribution partners [are] handled a lot of times in the workbenches now.”

Different Lines, Different Timelines

Wein says the implications will vary significantly by line of business. Specialty lines may be among the earliest to experience pressure on traditional core-system models because insurers need to launch products quickly, test and learn from emerging risks, collect different forms of data and turn quotes rapidly. In specialty insurance, he notes, a slow quote can mean a lost opportunity.

“The challenge with [specialty lines] is to spin up a lot of products and then test and learn off those products and then spin up new ones off of very unique risks that they’re covering,” Wein says. “You don’t want to spend six months and a million dollars to spin up a new product every time you want to put one out there in specialty, because you can’t do that. It’s not economic.”

By contrast, admitted lines such as workers’ compensation may not change as quickly, Wein suggests, because they are more regulated, prescriptive and controlled. In such lines, variations by state and compliance requirements may continue to support more traditional core-system approaches.

“Workers’ comp is a very controlled, very prescriptive line of business,” Wein says. “The pricing is controlled. The features of the products are controlled.”

Life and group insurance present a different pattern, according to Wein. In those markets, the policy administration system is only one part of a broader ecosystem involving benefit administration platforms, distribution partners, employer relationships and service integrations. That makes the architecture less about a single core platform and more about the ability to orchestrate relationships across an ecosystem.

“On the life side, the life side has always been a question of the ecosystem that surrounds the life PAS,” Wein says. “If we go to group, it’s always been about the integration with the benefit admins and with the partners.”

From Monoliths to Orchestrated Workflows

The result, Wein says, may be a gradual decomposition of large core-system categories into more modular, AI-orchestrated workflows. In that future, carriers may not replace one monolithic system with another. Instead, they may subscribe to or assemble workflows selectively, using AI to manage the complexity of many discrete processes running across multiple systems.

“Maybe it means that these big macro products decompose into agentic workflows and you pick and choose which ones you want in which context,” Wein says.

That approach would introduce its own complexity. A carrier might eventually have thousands of agentic workflows operating across underwriting, service, claims, billing, compliance and distribution.

“If I have 5,000 agentic workflows running around, that would be very complicated,” Wein says, summarizing the concern he hears from industry participants. “But then you’re going to have other agentic LLMs that are managing that.”

A Strategic Window for Vendors

Wein does not argue that today’s major core-system vendors will be displaced immediately. Rather, he suggests that the next several years may create a strategic window in which carriers and vendors must adapt to a faster-moving technology environment than the industry has historically been able to absorb.

The insurance industry has lived through earlier technology adoption cycles, including the emergence of the internet, e-commerce and cloud-based systems. Those changes unfolded over long periods. Wein says AI, and particularly agentic AI, is moving on a different clock.

“The technology will continue to evolve at an ever more exponential rate,” he says. “Only a few carriers are going to be able to adapt at that level. They’re not culturally set up to do it. They’re not funded to do it.”

That creates a gap, Wein suggests, between what technology can enable and what insurance organizations can realistically absorb. The vendors that serve insurers will have to navigate that gap while rethinking product strategy, implementation models and the long-term meaning of core systems.

“The challenge for the vendors is to reimagine the products in a different world,” Wein says. “A world that’s really just a little bit ahead of where we are right now.”

Wein compares the potential disruption to earlier moments when technology helped insurers or vendors change the basis of competition. Some of the companies that benefit most from the next phase of AI-enabled insurance architecture may not yet be widely known, he suggests.

“There’s probably a couple of companies out there that we don’t know their names right now that will be very, very large firms five or ten years out from here,” Wein says. “I don’t know who they are yet. But what we do know is the outlines of what they’re going to incorporate to be successful.”

Beyond Agentic AI

Wein also sees agentic AI as part of a broader technology wave that includes quantum computing, AI infrastructure optimization and more dynamic approaches to risk modeling. Those topics, he says, could eventually affect insurer cybersecurity, compute costs and the ability to model risk in real time. However, he characterizes MCP-enabled agentic workflows as the more immediate strategic issue for insurers and their technology providers.

The urgency, Wein suggests, lies in the possibility that core-system boundaries may change before many insurers have finished modernizing around the previous generation of assumptions.

Socotra CEO Warns Against ‘AI Demo Theater’

Combined Ratio Solutions Offers an Alternative Path for Core Systems