AI Agents Need Business Context to Make Better Decisions

AI agents can take actions inside a business, but having the right information is essential to making sound decisions. A supplier payment might pass checks against a purchase order, budget and supplier record, yet still be wrong if the contract expired the day before.

In a TechRadar article, Will McAllister, identified as Senior VP and Managing Director for EMEA at Guidewire, argues that enterprise AI needs more than capable models. Assistants that summarize information or make recommendations still rely on people to review their output, and that review can fail. Agents raise the stakes when they can update records, approve requests or trigger workflows directly.

McAllister says businesses should give agents visibility into current goals, rules and task information without granting them unrestricted authority. He calls for “broad context, but with narrow authority.” Feeding agents company documents piecemeal may leave them with outdated rules or examples that do not reflect how exceptions are handled.

His proposed approach, “context at the core,” places AI within operational platforms that record commitments, apply rules and handle transactions. A trusted core can bring together current data, rules, permissions and history, while agents retrieve the smallest set of authoritative facts needed for a task.

McAllister points to insurance as an example: decisions can depend on policy terms, customer circumstances, local rules and a claim’s current status. A document extract from the previous week may be out of date if a policy changed the day before. He says the same need for operational context applies to other complex enterprise processes.