FIELD NOTE 03 / HOW AI USES THE PRODUCT
From application interfaces to governed AI extensibility
The application remains important. Governed AI access becomes another interface.DIRECT ANSWER
The short version
An AI-extensible insurance platform exposes governed capabilities that customer-owned agents can discover, invoke, and combine. Beneath the screen are APIs, connectors, permissions, durable data models, tests, approvals, and audit trails. The customer can assemble a specific workflow without rebuilding the reliable core or giving an agent uncontrolled production access.What changes
The application may no longer be the only meaningful interface. A customer may also begin inside an enterprise assistant, coding agent, copilot, or internal system. Product strategy therefore includes whether those systems can understand the platform’s actions and safely compose them into customer-specific work.
What does not
Governance remains non-negotiable. Permissions, testing, human review, approvals, observability, and audit records define what an agent may do. Extensibility is not permissionless code generation inside production systems.
An illustrative insurance workflow
Hypothetical example, not a customer case study.
A carrier’s internal agent reads appetite, checks policy data, prepares a quote draft, and routes an exception to an underwriter. It invokes controlled platform actions rather than scraping a screen. The underwriter retains authority, and every action remains reviewable.
DOCUMENTED EXAMPLE
Guidewire: knowledge access is not production authority
Guidewire’s Qusar documentation describes a read-only MCP interface that gives a developer’s AI coding tool access to curated Guidewire knowledge. It separately describes an Agentic Framework for tool-using agents and multistep workflows. Helping AI understand a platform is not the same as authorizing it to change production data.
What this does not establish: Guidewire’s security guidance leaves application inputs, model-bound data, and custom-tool permissions with the implementing team. Published extension capabilities do not prove that every customer deployment is safe or effective.
- Guidewire Qusar platform release documentation ↗August 2026 release · Checked September 2, 2026
- Developing secure AI agents on Guidewire ↗Checked September 2, 2026
The Season 2 question
Season 2 asks which capabilities agents should reach, who governs them, and where AI access complements the application interface or influences platform selection.