Yesterday, at its Connections conference in Las Vegas, Guidewire launched a platform approach to agentic insurance — agents running on the same platform as its core applications, working with the insurer's current data and workflows — and, the same day, announced a definitive agreement to acquire hyperexponential, the agentic underwriting and pricing platform, for $1.7 billion in cash plus up to $400 million in contingent consideration. Guidewire says hyperexponential has powered underwriting decisions across more than $75 billion in annual commercial P&C premium.
Read those two announcements together and the story is not "another vendor has agents." It is that the agent layer is moving inside the core. The question is no longer whether to experiment with agents. It is what you still own once your core vendor ships them.
Guidewire's pitch is, on its face, the right one. Its controls enforce three boundaries: what agents know, what has already been done, and what they are allowed to do — with consistent security and a clear audit trail. UnderwritingCenter turns submissions into risk profiles matched to appetite and moves the quote into PolicyCenter, removing handoffs. Concierge lets autonomous agents capture loss details and deliver decision-ready claims to ClaimCenter. Its Agentic Framework lets insurers use Guidewire-built agents or build their own, choose the model per task, enforce guardrails, and monitor costs. That is governed work where the work already happens, not another bolted-on workbench.
What it means for carriers
First, convenience now has a price, and the price is exit. An agent inside your core, reading your core data and writing back into your core workflow, is faster to deploy than one you integrate yourself — and harder to leave. The mandate — what the agent may do, up to what limit, with whose approval, and how it is suspended — will increasingly live in vendor configuration rather than in a document you control. Model portability is not architectural portability: swapping the model is easy; lifting the mandate, the workflow, and the audit history out of the platform is not. The $1.7 billion price tag tells you how valuable that stickiness is. Price it the same way.
Second, the audit trail is now the vendor's audit trail, and that matters the day a regulator or claimant asks for it. Guidewire promises a clear trail that stays current through updates. Good — but whose tape is it? Can you export the full decision record for one policy or claim, replay it outside the platform, and show an examiner what the agent saw, what it recommended, and what the human confirmed? If the ledger only exists inside the system that produced the decision, you have observability, not accountability. Accountability means the tape survives the vendor relationship, the upgrade, and the dispute.
Third, differentiation moves up a level. When every carrier on the same core gets submission summaries, risk insights, and next-best actions in the same flow, the tool stops being the edge. The edge becomes appetite, data, and judgment: which risks you take, what your own book teaches you that the platform's models do not, and how cleanly your underwriters can explain a decision the agent helped make. That beats a features race — but only for carriers that defined appetite and evidence standards before switching agents on.
From the operator's chair
One: write the mandate outside the vendor before you enable a single core agent. One document, in your words, not the platform's: the entity the agent represents, its activities, its limits, the human approvals required, the evidence of authorized execution, and how the mandate is suspended or revoked. This is binding authority discipline applied to software — and an agent should never gain underwriting authority simply because another system accepts its technical credentials.
Two: run the ledger portability test before you run the pilot. Pick one claim and one submission, export the complete decision tape, and replay it somewhere the vendor does not control. If you cannot, you have found your first governance gap, and it is cheaper to close it now than in front of an examiner.
Three: rebuild one journey before you turn on the portfolio. Intake is the obvious candidate — where broker time, underwriter time, and quote-cost economics collide. Delete the steps that should not exist, group the AI work under a single human check, keep the tape — and keep at least one capability genuinely vendor-neutral, such as appetite or pricing behind a governed interface, so replacing a platform in three years means replacing a component, not rebuilding the company. The winners will not have the most agents. They will have agents that still answer to a mandate they wrote, a ledger they own, and a journey they actually rebuilt.
— Krishna