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

Speaker 2: Let's bring her in to the TBPN UltraDome if we can. Let's give it a try. Hello. Welcome to the show.

Speaker 10: Good morning. Thank you.

Speaker 2: Thank you so much for taking the time to join us. Please introduce yourself and tell us a little bit about what's going on today with AWS.

Speaker 10: Sure. So Colleen Aubrey. Been at Amazon for over twenty years now and joined AWS two years ago. Okay. And I've been on this journey of of looking at how can we bring to life some of the operational expertise that we've developed at Amazon Yeah. Into Agentyx products and put them in the hands of AWS customers. And today, it was really sort of great to announce two new products, Connect Decisions, Amazon Connect Decisions, Amazon Connect Talent. We announced Amazon Connect Health a couple of weeks ago, and this really expands the family of Connect products that we have adding to Amazon Connect customer, which is a nine year old product that we've been developing.

Speaker 2: So can you help me understand? I think most people will be familiar with AI chatbots and many people will be familiar with AI agents and coding agents that they fire off a prompt and maybe it goes for five, six, eight hours sometimes and then comes back with a fantastic response. But when you're deploying an advanced model, a frontier model, an agent into an into an enterprise system that's deployed on AWS, what are what are your clients looking for? How are they thinking about integrating an agent into a platform? Is it that they're vending the agentic workflow to their customers in turn? Can you walk me through some of the examples of how enterprises are actually deploying agents in practice when most people are probably familiar with the more point solution they type the prompt themselves?

Speaker 10: Yeah. Well, let me first by start by saying that I think we're early in that journey. Yeah. And, certainly, in my conversations with customers, a lot of what I hear is people looking to automate processes.

Speaker 2: Sure.

Speaker 10: And so and and for me, what that triggers is that the mental model is looking at how work gets done today Yeah. And then putting AI to work to do that same process. Totally. And personally, my point of view is that that misses the bigger opportunity. Mhmm. I think the bigger opportunity is much more transformative, and we can't assume that how we work today is the sort of gold standard for how we might work when we have this new capability in our in our hands, which is to be able to develop fairly complex agentic teammates Yeah. That can work alongside people in the business. So we're really going after this this mission of how do you develop an agentic teammate, which is in the business with people and is actually transforming how work gets done and and do that in a way which is intuitive and natural so that transformation doesn't come with heavyweight change management. And so we really have gone back to the drawing board and and 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.

Speaker 2: Sure. Sure. I I feel like AWS' strong suit has always been understanding complex systems, being able to deploy a ton of resources, but then also being able to monitor, understand everything from cost to performance. How is that changing in the age of AI agents? I imagine if I have a bunch of digital workers doing stuff, I want a water cooler where I can tap someone on the shoulder and say, hey, how's it going? I want analytics on how different work processes, even if they are fully automated, are continuing and progressing. What what does the analytics stack look like in the future?

Speaker 10: Yeah. I think it's a good question. And certainly, in the agentic teammates that we're building Yeah. There's a good amount of thought that we put into the observability Mhmm. The trustworthiness, like, an an AI's an agentic teammate's ability to explain why they've taken some action, why they've come to some conclusion, reasoning that they've gone through. And an example of that is in Amazon Connect Health. In this case, we're working on behalf of providers, preparing sort of summarization of a patient's history before they meet with the patient. And in that case, every reference, every conclusion is traceable into labs, into previous visits, into medications, so that the physician in this case versus builder or an SDE has the opportunity to be able to observe the work, to be able to reason over the work, to be able to also course correct as And so for me, I'm really much more thinking about how I make an agentic teammate trustworthy and part of the team, and that that observability comes to life in different ways. Of course, the covers, we have to solve for price performance. We have to solve for latency. We have to solve for managing the efficiency of deploying many agents under the cover to be actually be able to deliver the experience, and and that's something some of the hard work that we're doing on behalf of customers.

Speaker 2: Yeah. I I feel like it's potentially underappreciated, And maybe this is a side an outgrowth of the fact that the first AI models that most people interacted with were sort of low powered chatbots that did hallucinate. But the hallucination problem has, at least to my experience, basically gone away. And it feels like the next challenge is actually educating people on the reliability, the traceability that you mentioned. And of course, I'd be interested to hear a few different there's a whole bunch of different projects that you can embark on to build confidence around the traceability and reliability. I did have one last question though. I want since you've been with Amazon for twenty years, I'd love to know how has the company changed? How has it stayed the same?

Speaker 10: You know, I think, certainly for me, this sort of passion for inventing on behalf of customers

Speaker 2: Mhmm.

Speaker 10: Has remained pretty consistent through, like, the time that I've been at the company. And so really going deep to understand a customer problem, but then also pulling back and asking ourselves, is there a better way? Is there is can we invent a way of solving this problem which is meaningfully better than what exists today? And do we have a new unique point of view about how we would do that? And that's really what we're trying to do with this connect family of products is really, know, ask ourselves honestly, what is the unique capability, the unique operational experience that we have that we can bring to life? And I think that has remained consistent. Otherwise, you know, Amazon's always evolving, always changing. Yeah. And, you know, our our businesses go through inception. They go through scale. They go through reinvention. And now is a great time for all Amazon teams to be thinking about reinvention and how do they deliver their products and services to customers now that we have this new AI capability on our hands? And it's exciting to see the energy around that, and I feel sort of like sort of just freshness in the organization as we go through this next phase.

Speaker 2: That's great. Well, thank you for coming on the show, and congrats on the progress. We'll kick it over to Anthony next. Thank Have a for great rest of your day. We'll talk to you soon. Thank you. Up next, we have Anthony, who is a VP and distinguished engineer at AWS as well, to continue our segment on AWS and detail a little bit of the OpenAI integration and new stateful runtime for agentic systems. And so we have Anthony in the TBPN Ultra. Anthony, welcome to the show. How are you doing?