Meet hyperoperator: the AI agent for commercial P&C underwriting

Today at hxLive London, we launched hyperoperator: the first agent capable of completing complex underwriting tasks in commercial P&C.

It advances submissions from broker email to priced risk inside the appetite, pricing, and authority controls carriers already trust, coordinating the work that surrounds each underwriting decision.

Why we built it

Commercial P&C underwriters face pressure from all sides.

More submissions are arriving. Brokers expect faster responses. Rate adequacy is deteriorating as market conditions soften. Underwriting teams need to price the right risks, protect portfolio performance, and keep judgment consistent across teams, lines, and regions.

At the same time, the work around each decision has become harder to manage. Appetite lives in guidelines, emails, and exceptions. Pricing logic sits in models and engines. Portfolio context arrives too late, often after the decision has already been made. Authority rules vary by stage and line of business. Audit requirements follow every consequential call.

Generic AI can read a submission. It can extract data. It can summarize a document.

That is not enough for live underwriting.

To operate safely in commercial P&C, an agent needs the full underwriting context: appetite, pricing logic, workflow rules, portfolio signals, authority controls, and a decision trace. It needs to know when to act, when to pause, and what the underwriter needs for review.

That is what hyperoperator does.

Four modes of agent-led underwriting

Carriers configure hyperoperator around their own rules, confidence thresholds, and authority model. The same agent can work in four modes, depending on the submission, line of business, and level of human review required.

1. Straight-through execution

For routine work inside defined thresholds, hyperoperator advances the submission without waiting for an underwriter to push each step forward.

It can ingest the broker email, structure the submission data, run clearance, apply appetite logic, trigger the right pricing workflow, and prepare the file for decision. The work moves through the carrier's own rules, not a separate AI process running alongside them.

If the submission meets appropriate criteria - for example renewals with minimal changes and standard rate change guidance that has been signed off by underwriting - this can even mean hyperoperator being able to take the submission all the way through to quote without an underwriter intervening.

The point is not autonomy for its own sake. It is controlled execution where the rules are clear and the risk sits within authority.

2. Human-in-the-loop review

Not every underwriting decision should move straight through. Many should not.

When a referral, exception, pricing judgment, or authority threshold requires human review, hyperoperator pauses the work and brings the underwriter the context needed to decide. It extracts the data, prepares the pricing view, and surfaces the relevant appetite rules, portfolio signals, and exceptions.

The underwriter does not start from a blank screen or a pile of attachments.

AI prepares the case. AI surfaces the relevant information. Underwriters make the call.

3. Ad hoc underwriting analysis

Underwriters also need to ask questions as the work moves.

What would we have priced this risk at six months ago? Is this account inside appetite for the current portfolio strategy? How does this submission compare with similar risks already on the book? What happens if we adjust the deductible or attachment point?

hyperoperator lets underwriters ask those questions inside the underwriting workflow. The answers draw on the carrier's pricing logic, portfolio context, and decision history inside hx, rather than a generic model response disconnected from the workflow.

4. Always-on portfolio monitoring

Some underwriting work happens outside individual submissions.

Portfolio concentration builds. Adequacy drifts. New exposure patterns emerge. Small exceptions become trends only visible at book level.

hyperoperator can monitor those signals continuously and surface what needs attention. It can flag anomalies, identify areas where appetite may need review, and bring portfolio context back into live underwriting decisions.

That changes the rhythm of portfolio steering. Leaders do not have to wait for a delayed report to see how the book is changing. Portfolio signals arrive while there is still time to act.

What makes hyperoperator possible

Frontier models and general purpose agents are powerful for knowledge work, but they lack the domain understanding, the decision infrastructure, and the guardrails to meet the needs of real-world underwriting. What enables hyperoperator to support the end-to-end underwriting journey is the depth of the hx platform underneath, proven on $75bn+ in annual commercial P&C premium.

hyperoperator runs in a secure, governed environment on a carrier's private cloud. It is steered by hx's Workflow Builder, calls the platform's Calculation Engines and specialist agents as tools, and relies on Portfolio Intelligence for full context of the carrier's book.

Environment — Workflow Builder. Each carrier defines the lines, stages, handoffs, and review points hyperoperator operates within. It works inside the workflows the carrier has configured in hx, not outside them.

Tools — Calculation Engines and specialist agents. hyperoperator calls deterministic Calculation Engines to apply pricing, rating, appetite, referral, and authority logic in a scalable, auditable framework. It can also delegate focused tasks — such as data ingestion, risk analysis, deep research, and model deployment — to specialist agents.

Context — Portfolio Intelligence and underwriting memory. hyperoperator sees the submission in front of it, but also the broader context around it: underwriting guidelines, business logic, underwriter notes, decision traces, in-the-moment judgments, and portfolio-level signals such as exposure, concentration, and adequacy drift.

Agentic underwriting cannot work without this operating layer: deterministic workflows, integrations, memory, testing frameworks, observability, deployment, and governance, which combine to unlock real transformation.

That is the difference between an agent that stops at ingesting a broker submission and an agent that can truly advance underwriting work as part of a wider agentic underwriting workbench.

Watch the keynote highlights

You can watch here the highlights from the hx Live keynote address.