Let AI Ask the Compliance Question, Keep the Verdict Deterministic
AWS introduces a design pattern it calls Adjudicated Query for high-stakes compliance work: business users ask their compliance questions in a chat interface, Amazon Quick, while the actual pass/fail decisions stay in a deterministic, non-AI rules engine.
The split is the point. The conversational layer handles a question typed in ordinary language; the verdict stays in code that is meant to be reproducible and auditable. It is a tidy division of labour for domains where a fluent answer is not the same thing as a defensible one.
AI-generated discussion · Different perspectives
The scale is what makes the question worth asking: a portfolio operator with 50,000 leases spread across multiple states, each publishing landlord-tenant statutes — late-fee caps, notice periods, security-deposit limits — that change on the legislature's schedule, not the operator's.
Chat is a forgiving interface; a compliance verdict is not. That is the real design decision here — not which model writes the answer, but which system is allowed to be wrong.