AWS launches Amazon Bedrock Managed Agents with OpenAI integration, targeting enterprise agentic workflows

Colleen Aubrey has spent over twenty years at Amazon and joined AWS two years ago with a specific mandate: take the operational expertise Amazon built internally and turn it into agentic products for enterprise customers.

The announcement she's here to discuss is the expansion of the Amazon Connect family. AWS launched Amazon Connect Decisions and Amazon Connect Talent, joining Amazon Connect Health, which was announced weeks earlier. These sit alongside Amazon Connect Customer, a nine-year-old product, and together form a suite of agentic teammates targeting distinct business functions — decision-making and supply/demand planning, healthcare, recruiting, and customer journeys.

“We really have gone back to the drawing board trying to think about how will work happen in a number of different areas — in decision making within supply and demand planning, in health care, in recruiting, and in customer journeys... Today, it was great to announce two new products: Amazon Connect Decisions, Amazon Connect Talent.”

What "agentic teammate" actually means

Aubrey draws a clear line between the obvious use case and the one she thinks matters. Most enterprises she talks to want to automate existing processes — take what workers do today and hand it to AI. Her view is that this approach leaves most of the value on the table. The bigger opportunity is redesigning how work happens, not just who does it, and doing it in a way that doesn't require heavy change management.

How trust and observability get built in

Amazon Connect Health is the clearest example she gives. The product works on behalf of healthcare providers, generating patient history summaries before physician visits. Every conclusion — every reference to a lab result, prior visit, or medication — is traceable back to the source. The physician can observe the reasoning, interrogate it, and correct it. Aubrey frames this as the model for what a trustworthy agentic teammate looks like: explainable by default, not explainable on request.

Under the hood, AWS is also managing price performance, latency, and the orchestration of multiple agents running in parallel — work she describes as being done on behalf of customers rather than exposed to them.

The hallucination question

Aubrey largely agrees that hallucination, as most enterprise buyers experienced it in early chatbot deployments, is no longer the central problem. The harder challenge now is convincing organizations that traceability and reliability are real — and building the workflows that demonstrate it in practice.

The Amazon constant

Asked how Amazon has changed over two decades, Aubrey's answer is that the core habit hasn't: go deep on a customer problem, then pull back and ask whether there's a meaningfully better solution, not just a faster version of the existing one. The Connect family is her attempt to apply that discipline to agentic AI — starting from what AWS knows how to do uniquely, rather than chasing the generic workflow automation everyone else is building.