Production-grade AI agents for financial compliance: Lessons from Stripe
TL;DR
Stripe describes a compliance agent system on AWS Bedrock that supports human reviewers in financial crime reviews, while keeping final decisions with experts. The system breaks complex reviews into smaller sub-questions arranged as a DAG. Agent outputs are used as supplemental research, and human-validated answers feed later questions.
Nauti's Take
The useful lesson is not the Bedrock gloss, it is the role split: agents gather evidence, humans decide. That is how you put AI into regulated workflows: small questions, hard logs, clear accountability.
Agents without audit trails just accelerate liability.
Briefingshow
The important part is not that Stripe built an agent, but where it put the boundaries. The architecture treats compliance as a controlled workflow, not a chatbot problem, with logs, sub-tasks, human accountability, and infrastructure for long-running, expensive agent jobs. That is where production AI separates from demo automation.
Sources
