How to onboard a new underwriter to an AI agent workspace

Onboard a new underwriter by separating two things the agent makes easy to blur: assembling a risk and accepting it. In the first week the newcomer checks the agent's submission summaries against the source documents. In the second they price and refer under a senior's countersignature. Binding authority comes later, matched to their own certified limits.

What does the agent do in underwriting, and what does the underwriter keep?

In a typical underwriting team the agents do the assembly. An intake agent reads a broker submission, extracts the schedule, the loss history and the declared values, and flags missing items, at depth 1 or 2. A risk agent drafts a summary against the underwriting guide and proposes a referral where the risk falls outside appetite. A quoting agent may draft terms for a person to issue.

The underwriter keeps the judgement: whether the risk is acceptable, at what price, on what terms, and whether to bind. Binding is hard to undo, so where it is automated at all it sits at depth 4, and most teams keep it with a person.

The underwriting manager owns the onboarding. A senior underwriter with authority above the newcomer's is the buddy. How the daily work shifts is covered in what changes in an underwriter's day when AI agents arrive.

The onboarding steps

1. Day 1. Person (underwriting manager): grants read access to the workspace and a practice space. A sandbox workspace with past submissions lets the newcomer explore without touching live risks.
2. Days 1 to 3. Agent, depth 1 to 2: extracts and summarises live submissions; the newcomer checks five summaries a day against the broker's documents. They look for the gap between what the broker declared and what the agent wrote. A declared value of 2 million appearing in the summary as a fact, not as a broker's declaration, is the kind of slip to catch.
3. Day 4. Person (newcomer): reads the provenance on every fact in two summaries. Loss runs from an insurer's system arrive signed or sealed; values stated by the insured are declared; a survey report attached to an email is external until verified.
4. Day 5. Person (newcomer and buddy): walk the referral rules. Which risks the agent refers and why. A trade outside appetite refused at the scope bound. A sum insured above the team's limit held for a person with higher authority.
5. Days 6 to 8. Agent, depth 2: drafts pricing notes and terms; person (newcomer) prices, buddy countersigns. Start with renewals of clean risks before new business.
6. Day 9. Person (newcomer): records disagreements with the agent's summary and fixes the cause. If the agent keeps missing a policy exclusion the guide added, the guide source is stale.
7. Day 10. Person (underwriting manager): reviews the record and sets the newcomer's own authority, usually a lower limit than the buddy's, with binding above it held for a senior.

Which controls protect the book while the newcomer learns?

Authority limits held by the workspace. The newcomer's limit is part of the business's own limits. A quote above it is held for a senior. Evidence: the verdict names the bound and the person who released it.

Countersignature on all pricing in week two. The record carries both names, which matters when a claim arrives two years later and someone asks who accepted the risk.

Provenance on every fact in the summary. Declared values are flagged as declared. An external survey cannot by itself clear a referral.

Practice before live work. Time in the sandbox leaves its own trail, so the manager can see what the newcomer explored before going live.

Where pricing affects people directly, regulation may add duties. The EU AI Act lists risk assessment and pricing for life and health insurance as high-risk. This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.

How do you know the new underwriter is ready?

• They catch summary errors the buddy would catch, and can say which source each fact came from.
• Their prices in week two sit close to the buddy's, and differences are reasoned, not random.
• They can explain every referral the agent made on their risks that week.
• They have fixed at least one cause, not just corrected one draft.

Where does underwriting onboarding go wrong?

Reading the summary instead of the submission. A clean, well-ordered summary invites trust. Underwriters who skip the broker's own documents miss what the agent left out, and the agent can only summarise what it was given.

Inheriting the buddy's authority by accident. If the newcomer is set up with the same rights as the senior they shadow, the workspace will not hold quotes it should hold. Set the newcomer's limit before day 6.

Never seeing a declined risk. Newcomers should review refusals as well as acceptances, so they know where the agent's scope ends. Frontline roles face a similar adjustment; compare onboarding a new customer support agent and onboarding a new collections agent. The overall method is in training teams for AI agents.

Where the Academy fits

Every lab in Pryme Intelligence Academy runs in a free Build workspace, so a new underwriter can practise reading verdicts and provenance before the live book. The Operator track's modules on how a governed agent decides and on human review match the first fortnight, and module tests are set in the learner's own industry and role.