The core system ships the agent

September was the month the incumbents moved. Sapiens, whose policy administration system runs a large share of the world's insurers, announced SapiensAIP, an AI-native platform that embeds agentic tools directly into the core system across underwriting, policy administration, billing, claims and customer engagement. It is built in three layers, an experience layer, an intelligence layer where the agents do the work, and a foundation layer drawn from the company's forty years of insurance data, and it ships with agent swarms that profile, map, validate and extract data for a migration, and a configuration hub that reads documents in any format and maps them to the right fields. Continental General was named as the first customer evaluating it.

When the system of record grows its own agent, the co-pilot sold into it becomes a feature. The company that is the function keeps its price.

Oracle Health did the same on the hospital side. Its Clinical AI Agent, expanded in August to cover coding, dictation and chart review, now extends to professional fee coding: it analyses the visit and suggests the charge codes inside the ordering workflow, and on 23 September Oracle announced a set of revenue cycle capabilities meant to catch reimbursement problems at the start of the patient and payment journey rather than at the end.

The independents moved too, and the contrast is the point. Sixfold launched an AI Underwriter for life and health that reads the medical and financial evidence, applies the carrier's own manual, appetite and authority levels, and returns a case-level decision to rate, refer, decline or postpone with its rationale attached; the company reports customers cutting case evaluation time by more than half and writing about thirty percent more premium per underwriter. Maverick Medical AI released what it calls a clinical-to-revenue foundation model, one system that runs documentation, coding, compliance and denial prevention through to billing-ready codes.

Read together, the month sorts the category. A tool that sells assistance into a workflow owned by someone else has just watched that owner ship the assistant natively, at no marginal price, to every customer at once. A company that runs the function end to end, holds the decision and its rationale, and prices against the underwriter's or coder's payroll rather than a seat, has something the core vendor cannot ship in a release: the liability, the license and the record of every decision it has made.

The regulatory clock reinforces it. The EU AI Act's obligations for high-risk systems took effect in August, and the NAIC's model bulletin on insurers' use of AI has now been adopted by roughly two dozen US states, each requiring a written governance program. A governed agent with an audit trail is not a feature. It is the product.