PrymeIntelligence

AI agent skills by role·Definition

What is exception handling?

Exception handling is how an operation deals with cases outside its normal rules. With AI agents, exceptions are held for people with the skill to decide them.

Exception handling is the way an operation deals with cases that fall outside its normal rules: an invoice that does not match its order, a customer who fails identity checks, a claim with no clear policy answer. With AI agents, routine cases flow through and exceptions are held or routed to people with the skill to decide them.

Why does exception handling become the main job?

Agents take on the routine work first. That is what depth 3 means: routine actions within bounds. What remains for people is disproportionately the hard part: mismatches, missing documents, unusual customers, conflicting rules. In many teams, the people who once processed a hundred routine items a day now decide a smaller number of exceptions that each need real thought.

That shift is a training issue. Processing routine work and judging exceptions are different skills. A clerk who could match invoices quickly may not know how to decide whether a supplier's changed bank details are genuine. If teams do not plan for the change, exceptions pile up or get waved through.

How does it work with an agent?

The agent recognises that a case is an exception, either because it breaks one of the customer's own limits or because a fact it needs is missing or weak. It then gives a verdict: held, with a draft and the reason, for a named person; or refused, with the bound that decided it. The case lands in an exception queue, where a reviewer sees the agent's draft, the facts with their provenance, and the reason it stopped.

The reviewer decides, corrects or escalates. The decision and reason are filed, and recurring exceptions are reviewed to see whether the rule, the agent or the process needs changing.

An example

Take a hypothetical bank's customer operations team. An agent processes address changes at depth 3. A customer asks to change their address and, in the same message, mentions they are leaving an abusive partner. The agent recognises the second point, holds the change, and routes it to the vulnerable customers team rather than processing it routinely. A specialist calls the customer, updates the address with extra protection on the account and records the steps taken. The vulnerable customer handling checklist covers that kind of case.

How is it different from complaint handling?

Complaint handling deals with a customer's expression of dissatisfaction: something has gone wrong, in their view, and they want it put right. Exception handling deals with cases that do not fit the rules, whether or not anyone is unhappy. A complaint can be routine (a late delivery with a standard remedy), and an exception can involve a satisfied customer (an unusually large but legitimate payment). They overlap when a complaint itself is unusual, but they are run as different processes.

What about exceptions to the agent's bounds?

A different kind of exception is a deliberate, temporary widening of an agent's bounds, for example allowing larger refunds during a service outage. Those should be rare, approved by the control owner, and set to end on a fixed date. The reasoning is in every exception to an agent's bounds should expire.

Related terms

Held action, exception queue, escalation path and red line all relate. The guide to AI agent skills by role shows how exception handling changes each role's work.

Where the Academy fits

Pryme Intelligence Academy's Operator track is built around the reviewer's side of exception handling: how a governed agent decides, why it held a case, how to read the facts and their provenance, and when to escalate. The practicum assesses two weeks of real review decisions, which in most teams means exceptions.

Questions people ask

Should every exception go to a person?

Every exception the agent is not entitled to resolve, yes. Some exceptions can be absorbed into the agent's routine work over time, once a rule is written and the agent is certified for it.

How do you know if exceptions are being handled well?

Check response times on the exception queue, sample decisions for accuracy, and watch for the same exception type recurring without anyone asking why.

Train the people who run your agents.

Pryme Intelligence Academy: free courses on the real product, audio lessons, labs in a free workspace, and certificates anyone can check on VerifyDoc.ai. Teams get seats, an admin view and a free readiness assessment.