AWS has launched Resolutions, an agentic operations service for handling high-volume work such as refund requests, transaction disputes, fraud investigations and insurance claims. Its notable pitch is that agents can carry cases through multiple steps, while people retain approval and exception handling.
AWS AI Watch analysis
What happened
AWS says managers can describe a process in plain language or upload standard operating procedures, then use Resolutions to build a Work plan. The service asks clarifying questions to define exceptions, decision rules and approval thresholds before applying that plan to incoming cases.
AI agents can gather information, apply policies, act across business applications and ask customers for missing details. AWS says each action and decision is recorded, with outcomes verified and progress visible to operations leaders. Work steps can use prebuilt agents, agents configured in Amazon Connect Customer or a business’s own agents.
For example, AWS describes a bank workflow that checks a transaction dispute against chargeback rules, requests a missing receipt and can issue a provisional credit. A human specialist reviews cases that need judgement. That is AWS’s description of the product’s intended operation, not evidence here of measured results in customer deployments.
Why it matters
This is a more consequential proposition than a chatbot answering routine questions: Resolutions is designed to take action across business systems and move a case towards completion. If it works as described, it could change how organisations handle repetitive work, not merely how quickly staff find information.
The dividing line is oversight. A useful system needs to know which steps it may take, when it must ask a customer for more information and when a person must decide. AWS says its Work plans make those rules explicit and its records make actions auditable. For businesses considering automation, those controls matter as much as the agent’s ability to complete the task.
Our read
The interesting part is the attempt to turn operating procedures into bounded workflows, with questions up front to expose the exceptions that documents tend to leave lurking in the margins. That is a sensible design target. It is not yet proof that complex cases will be resolved faster or more cheaply in practice, and the real test will be whether the human hand-offs and audit trail hold up when a case stops being routine.
What to watch
- Which workflows customers put into production, and what results they report.
- How Work plans handle exceptions and approval thresholds in real cases.
- Whether the audit records give specialists enough context to review decisions.
Discussion spark: Would you let AI agents take steps such as requesting documents or issuing provisional credits under set rules, or should those actions always wait for a human?
Sources and evidence
- Amazon Connect launches Resolutions, an agentic solution for high-volume operational work (19 November 2026, 20:06 UTC)
- Amazon Connect Customer now provides AI chat summaries in the CCP (19 November 2026, 20:06 UTC)
not affiliated with or endorsed by Amazon Web Services (AWS)