Specialised agents for the four reviews inside one MLR review
Read the full post on AWS BlogNothing a pharmaceutical company publishes, whether an advertisement, a brochure or a sales aid, reaches anyone until it has cleared medical, legal and regulatory review. Today that means cross-functional teams reading documents line by line against dozens of guidelines and hundreds of scientific sources. A single asset takes around 15 days on average, and large organizations run thousands of these every year.
The pressure to go faster isn't abstract. In September 2025 the FDA issued over 100 cease-and-desist letters and thousands of warning letters over misleading promotional content, one of its most concentrated enforcement actions in decades. Speeding the process up while making it less careful isn't an option.
What makes MLR review hard to automate is that it isn't really one review. Reviewers weigh the same document against the scientific references behind its claims, against regulatory codes such as the IFPMA Code of Practice, against branding guides, and against approved product information. Each calls for different expertise, and a medical reviewer and a legal reviewer are not looking for the same things.
So the system mirrors that split rather than flattening it. Specialised agents on Amazon Bedrock AgentCore, built with the Strands Agents SDK, each own one dimension and run in parallel, citing sources and proposing fixes rather than issuing verdicts. The reviewer stays the decision-maker and starts from a marked-up document instead of a blank one. Studies put the reduction in review time at up to 90%, and Accelerate Medical Content Review walks through a deployable sample.
