What's Next with AWS - AWS and OpenAI leaders on Agentic AI | Amazon Web Services

Please welcome the Chief Marketing Officer and Vice President of Worldwide Marketing for AWS, Julia White.

Hello and welcome to What's Next with AWS. Thank you so much for joining us here in the room and of course, across the world on the live stream as well. Now it is very clear that we are in the era of AI. And while it's certainly still early, it is and evolving quickly. It's becoming abundantly clear that agents will have a profound effect on how we all work. Now, today, we've got a compelling set of conversations to have on this topic, and of course, several exciting announcements. So without further ado, we're going to go ahead and jump right into it. So to share off the top with his very unique perspective, please join me in welcoming the CEO of AWS, Matt Garman. Welcome, Matt.

Awesome. All right. Now back in December, all the way back in December, if you remember that I was looking up and you talked about heading into a world of having a billion agents. So curious. It's been almost six months since then. Like, how do you feel? Do you feel like that's still true? Things changed.

So still on track. A billion or more we are. We'll get there. Exactly. I know one of the things is we've been working with agents more and more, obviously. I think the thing we've seen is that it really fundamentally changes the constraints, right? And so much of business has been about managing constraints. But when agents can run 24 over seven, they can work on behalf, they can work in the background. It changes those constraints. And so I was interested, if you know, how do you think about now, like with that, what becomes possible that wasn't possible before? Just thinking about how to run the business with different kinds of constraints.

Yeah. You know, in my area, right. I was looking at the average Amazon advertising area and usually like an ad buy would be 100, $200,000, take six months to get the creative. And now with their new agent, I was looking at, actually now comes down to like the point where small businesses have like a 300% growth in small business adoption because now it's so much more capable. Again, those constraints of what used to be hard and expensive are no longer the case anymore.

I love the, the theme on like, what's possible when the growth, right. Which is great. But I also, you know, we both read the headlines and there's certainly a lot of focus on AI agents and how they're replace jobs and what it means for people in their jobs. And I know that can sound in moments hyperbolic, right? But there's a lot of energy on this topic. So I want to just you to hit it head on. Like, what's your perspective? What do you see?

Yeah. When I was looking at the U.S. Bureau of Labor Statistics, right, about tech jobs, as you said, and it looks like they're over adding over 300,000 tech jobs and growing at two times the rate of other fields. So to your point, like, and then also I did, I did do a fact check across Amazon. We're on track to hire over 11,000 SDE interns in early career. So I know you get that question.

So now you get the chance at your at your seat to look across all things happening at Amazon and curious, what are you seeing across the company? Because we're obviously an operating company and a tech company and how we're using agents and what it's how it's helping and changing Amazon.

Yeah. When I think across, you know, whether it's Prime or even one of the things I saw was we overnight took the entire intranet, you know, almost a million, over a million employees. Right? And we just turned it into an authentic first experience. Yep. And it took off. And people are like, so happy and so delighted. And it's like a simple thing, but it changes the employee experience and, you know, the types of scale.

All right. Now, in terms of what we've been building and obviously we've been building with working with the genetic technology for over a year now. And you get to also see what like lots of customers are doing. So yeah, I feel like we kind of have a front row seat. So I want to share with others like what learnings, maybe what mistakes, what, what's just evolved. That's just different now as kind of as the years gone on.

Right? And I see that as well. Like it's common to see, and maybe it's a good first step to just infuse agents into what you're doing. But ultimately, that's not where you get that 510 X.

All right. I want to stay on that point. So there's a lot of headlines. There's analyst reports right around what agents mean for SaaS, B2B SaaS applications particularly. And actually we've seen market valuations change based on some of those proclamations. So you obviously have a very unique perspective knowing that so many of the SaaS applications run on AWS. And we also deliver our own solutions too. And so what do you think? Like, where are the headlines, right? Where are they not? Right?

Agreed. And good to see them moving fast and adapting in that way. All right, now let's get into some of the announcements we're going to talk about today. The first one is something that, honestly, I've literally never seen this much like excitement and energy from the early adopters. And that's our new Amazon Quick desktop experience that will be available in preview today. So very excited about that. No, Matt. I know you've been using it to you've been one of our early adopters, so you have first hand experience. But like what, what makes Amazon Quick this new experience such a game changer in your view?

Yes, absolutely authentic enthusiasm across every role that's worked on it. So on that note, we're going to shift to talk more about Amazon Quick and show it. So thank you, Matt, for your insights and we'll see you again in a little bit.

All right. Now our next guest has been helping us find ways to re:invent day to day productivity through agents. So to come tell us more, please welcome Jigar Thakkar, the Vice President of Amazon Quick.

Thank you Julia, it's great to be here. The way we work is not working. Every day you lose hours to work that actually does not need you there. Searching the summarizing the context. Switching Amazon Quick is built to give you all that time back. Today we are launching new technology that will change how work gets done. Your local files, calendar, email, other apps all coming together with Quick and the ability to create intelligent documents that connect with live data, update automatically and serve as interactive dashboards for any business function, whether it's a sales leader's pipeline review, pulling real time Salesforce data, or finance team's monthly close page aggregating ERP, SSL, and other internal systems. Set it up in minutes, and by end of the day, you'll wonder how you ever worked without it. Say you want to set up a meeting with your colleague on a new product launch. Normally, there's a few messages back and forth on Slack to check their schedule, then over to Outlook to send them the invite. Quick handles all of that with a single prompt, Quick knew the project, figured out the time zone, cross checked the calendars, sent the invite, and pinged him on Slack. No tab switching, no back and forth. So how did Quick know all of that? This is the knowledge graph in action. Quick connects to your systems. You already use your email, Slack, your CRM calendar, local documents and files. It pulls knowledge from all of them and builds a living map of your work. Quick connects the dots between people, projects, decisions, and actions. Quick knows you wherever your work happens. Quick knows what's relevant. It brings the most important items from across my channels to my attention. It has highlighted that one of my priority items today is preparing for a customer meeting I have this afternoon. Quick helps me with the prep as well. Quick performs deep research on the company and the industry. It pulls customer examples from our internal systems, graphs, our current roadmap, product roadmap from a file on my machine and layers in the briefing from the account team, the emails and the Slack threads. Using this knowledge, it creates a presentation I can use in my customer meeting. Hours of work done in minutes with Quick. We often need different assets for different tasks and Quick can help with all of it. Say we need this as a document. Or an email that summarizes the brief to the account team. Or a spreadsheet. Quick and build it. Everything you've seen so far is about my day, but I don't work alone. Teams I work with have a critical processes spread across siloed systems, and usually those 1 or 2 people who really know how the process works Quick can help you here too, with a new capability that enables teams to create dedicated collaboration spaces. For example, my finance team was tracking contract reviews across email threads or shared drive and approval system. Nothing connected. They can now use Quick to build a single workspace where the contract, the supporting documents from OneDrive and the approval status all live side by side. What used to require hopping between many systems now happens in one persistent view that stays current, and they're able to share a read only view with their stakeholders. To self-serve, they can submit new contracts and check on status of submissions. No more chasing people down for updates. This took minutes to build and Quick. Let's take a step back. You seen Quick schedule a meeting from a single prompt, prepared a customer briefing across six different systems, and generate artifacts in minutes. That's personal productivity. You have seen Quick bring together siloed systems into a single workspace that keeps an entire team on the same page. That's team productivity. And when everyone across an organization is working this way, that's enterprise productivity. So that's Quick. My team and I are very excited to see what will you do with it.

Thank you, thank you. Jigar. Thank you. Thanks for the demo. Appreciate it. I hope we got a sense of how it works. Now, obviously, we're working in the productivity space for a while with AI. I've been really helpful in a more reactive way, but we're turning the corner to something different. So tell me about what's driving that shift.

So, Julia, there's been an inflection point over the last few years. There have been these siloed bots where they're operating on fragmented old ways of doing things and old apps, and that doesn't work. So what we did was build Quick from the ground up. It knows you. It knows the full context of the people you work with. So you don't have to go to every application. All that context come to you.

And now you talked about that in the very like personalized experience that knows you like how, what is the context? Tell me a little bit more about how that works.

So every time you use Quick, it learns from you. It is collecting the documents, you deal with the project deadlines you have, the people you must interact with, and the escalations from a customer that just showed up. What's happening urgently in an email and Slack. It has all this information and that provides the context it needs to give you very relevant answers.

Got it. All right. What I have to, you know, for all the CIOs or other folks listening, what about the controls, access controls, identity management, privacy, all of that, right? If has it has ability to connect with all those, how do they trust that to work?

Well, yeah. So we have 20 years of heritage of AWS with the standards of quality, performance and. Compliance, security and privacy of all the users, all of that entire stack that Amazon AWS is built on is what Quicksight stands for.

Got it. So it's bringing all of that kind of the robustness of the AWS platform and security kind of the robustness of the AWS platform and security kind of the robustness of the AWS platform and security controls into how Quick is implemented, which is very important. Now, curious what obviously, I know inside the company is very excited about internal, but what are you hearing from our customers who are using Amazon Quick?

We have so many customers who are giving us feedback, and it's been really amazing. From BMW or 3M or Mondelez and Allianz and so many others. And over the last few months, what we're hearing is how they are cutting down production time for so many of these processes by almost 80%. In some examples, we are hearing how they cut down the time it takes to process customer issues more than 50%. So there are so many of these anecdotes and examples we hear. And one of my favorite one is when one of our customers started our desktop app, they said, you know, I've been waiting for something like this and this just feels magical, right?

That's always the best word, right? When technology is great, you know, it has that magical feel. Well, hey, thank you, Jigar, for showing us what Quick looks like and giving a sense of kind of how it works in the backdrop as well. I think, again, the best thing to do is try it. I have to say, trying to describe it is hard, but using it is has that magical feel. So thank you so much for joining us and showing us about Quick.

Thank you. Take care.

All right. Now, as we have worked with agents, we become more convinced that it has the capability to transform, obviously, individual productivity, but actually entire business functions. So our next guest is a leader who has been at the very forefront of inventing and operating across several businesses at Amazon for over 20 years. So to share how we're applying agents to new business areas, please welcome Colleen Aubrey.

Welcome. Good morning.

Good morning. Thanks for joining.

Yeah. Of course.

All right. We've got a we're warming up here. All right. Now I just want to start with, you know, you've been worked across for 20 years, the retail business for a long time. And you're one of the founders of Amazon advertising business. And now joining AWS, leading our applied AI solutions area. So I first want to start with what are you and your team do? What is this area about?

Yeah. So I've had the great privilege of working with many of our consumer businesses. And really in that process, learned a lot about how we run businesses at scale at Amazon and how do we go through those different phases of development, from an early idea to finding fit, to actually operating at this, you know, at the scale of, of hundreds of millions of customers and billions of dollars. And so that experience really led me to AWS two years ago, joined the team after working on the ads business for a long time. And for me, there was this opportunity to take what we've learned in our business. And in many cases, we sort of outgrow off the shelf applications, and we end up building our own applications to run our business, usually around scale or complexity, pace of innovation. And so it's really for me, like we've learned a lot doing that. And surely there's an opportunity for us to take that learning, take those capabilities and put them in the hands of AWS customers. And we have this incredible sort of collision of opportunities with AI and this new sort of tool in the toolbox to think about like, what? How would we work now that we have this new capabilities? And so that led to applied AI solutions now.

And I know not just you, but many of your leaders actually came from the operating side of Amazon as well, kind of bringing in that domain expertise of across how Amazon works, right?

Yeah. And so it was important for me, like, you know, I'm sure many of you have seen this as well. It's often the mix of capabilities you have in the organization that brings fresh ideas. And so one of the important factors with me for me was like building a leadership team that have been in the business of being hands on, running at scale, delivering for customers every day, managing the complexity of doing that and the, the emerging sort of adaptability that you need to, to make that happen. And as part of that, you know, and my experience and many of these leaders experience at Amazon, we've had to build our own applications to run. And so for me, this mix of people in the organization is about it's not about SaaS for SaaS at stake. It's about, you know, inventing applications that work in the business and bringing that experience to life, which I think is an important, important thing.

We'll dig in more deeply. But your team is also responsible for Amazon Connect, right, which is our solution for customer experience and contact center. And so those of you not familiar with Amazon Connect, it's a leading enterprise customer experience application, and it's reached over $1 billion in revenue last year. But I think for maybe most important for today's conversation is you are actively taking that what is maybe had been a traditional SaaS application and you're making it an agent first and agentic experience. And so, you know, as Matt and I talked about, this is what all SaaS needs to do. So how are you doing it? What does it look like from the inside?

Yeah. So lucky for me when I joined AWS, actually, you know, Andy and Pasquale had already done started down this path of taking what we had built for our own stores, customer service team and bringing that to life in Amazon Connect. So last year we did 20 million interactions a day. We did a 12 billion of AI minutes and conversations last year alone. And so that products in its ninth year, in March of last year, we introduced what we think of as the next generation of Connect. And the reason why we think of it as next generation is we have introduced AI across the entire customer journey, voice or digital channels into the human experience, like assisting agents as they're working with customers into transcription, sentiment analysis, into the analytics and into outbound communication. And so the for me, the key here is that if you, if you like, put AI in a box and if you like, constrain it in that way, you're really not likely to realize the potential. And it's really when you can carry context across a journey. And we start thinking about now about the horizontal journey of a customer through, as they work with the organization that the power becomes, in my opinion, more real and more transformative. And so here we sort of launched Connect with unlimited AI. And, you know, I think it's sort of it's a journey, but I'm really pleased about the way we're really sort of putting agentic and AI context across the whole experience into really the teams that are coming back and rethinking it.

And I love that you take the whole journey, right? Not just parts of it and rethink of what's possible now with agents in that experience. Now we're taking that agent first approach and marrying it with Amazon's operating and domain experience to build more agentic solutions. So I'd love for you to talk about what we're announcing today.

Yeah, it's sort of a big day for, for my team and for me, like we've been toiling away for a couple of years ago, really challenging ourselves to rethink. And so today we have a few new products to share with you. And where we're going is actually, we are going to expand the Amazon Connect name to be a family of products. And so it's super exciting that the birth of a new family. And so we are unveiling three new agent core solutions Amazon Connect Decisions, Amazon Connect Talent and Amazon Connect Health. And the product that we've come to know and love is Amazon. Amazon Connect will be Amazon Connect Customer. And so this is our, you know, I guess for me, Amazon Connect starts to stand for agentic business solutions and bringing all that domain expertise together with the agent first.

Now let's just start for a second before we jump into each one of those. But what makes these energetic solution versus just kind of AI powered?

Yeah. You know, I think Matt talked a little bit about this when we thought about developing these products. The question is like, now that you we can see the future of what agentic capabilities can do, how do you think about how companies will work? And at the heart of it, we got to this idea of we need to build agentic teammates, and no one needs a difficult teammate. That's not great. Plenty of those. Yeah. So we don't need to do that. So where we need to go is really I need a teammate, which is easy to work with, intuitive. It's trustworthy and agentic team that is always learning. And I need to be able to take this agentic capability and get the transformation in my business, the reinvention of how we work. But without the change management, no one needs to go on a 2 or 3 year change management sort of process. And so we really are challenging ourselves to put agentic capabilities to work in a way that delivers that organization's new ways of working, but also in a very intuitive and easy way that I don't have to really learn something different. And we have developed a new device design philosophy that came with that. What we realized is we dug into this is that great AI with a underwhelming user experience is probably not going to be transformative that we really needed to address both. And so we came up with this, this sort of philosophy, which we call morphism. And I think in the past, in design, in design, you really went with skeuomorphic skeuomorphism, like with Apple, with the leaders here, where you actually looked at the physics of objects and translating that to the digital environment from a desk to a desktop, files to folders, those sorts of things. What we're focused on with Hugh morphism is what is the dynamics of human interaction? If we're building products that are at the heart of which is an agentic teammate, then how should those teammates interact with you? And so it's a journey. You know, we're getting started, but this has really been a key part of how we thought about these products.

So here kind of a mental model of this is what makes it agentic first, it's like a teammate, right? That's a different way to think about it. And then you literally have come up with another design approach for doing these as well.

Yeah, I would say it's work in progress. We're making good progress and you'll see some of it today. But this is like a journey and we've got a lot to unpack. And like every week, you know, Hector and his team are like exploring new interactions. And we're sort of pushing ourselves to think of like, is that really, is that how it happens? Is that awkward? Is that weird? Do I trust that? How do I think about how natural that is? And so it's, it's something that we're working through.

Well, interesting and excited to watch it happen. But let's get into a specific example. And we're going to start with an area that Amazon is quite well known for, which is rapid decision making, but also particularly in the area of supply chain and logistics, obviously from our retail store and your experience and the retail side, you've seen quite a bit of that evolution from the inside. So I want to start with what is it around Amazon's expertise that we're bringing into this domain?

Yeah. So let me start with you. Something we talked about. We've talked about a lot at Amazon for a long time. But speed matters in business and specifically high quality but timely and fast decision making is key to, you know, continually to innovate re:invent and to deliver for customers. And so this is sort of at the heart of, you know, heart of the product here and where we know there's one area that we do this daily, and it's complicated and it's something we spend a lot of time on is around managing demand and supply planning. And so we have decided to start there. We've built a product which we have two teammates here. One is a demand side planner and one is a supply side planner. They sit side by side. And this is not about data processing. It's not about delivering you another pretty dashboard. It's really about getting that high quality decisions at the speed of today's business, moving planners from reactive to strategic decision making, getting to the point where they're actually managing the business versus reacting to the escalations and the external conditions.

Yeah. All right. I'm going to invite Olivia to come out here in a second to show us what Amazon Connect Decisions looks like. But before that, what is it like? What about Amazon's domain expertise in this area specifically that's brought into this experience?

Yeah. So decisions as speed is fundamental to our stores business. And you know, we've been working on this for 30 years, building the capabilities to, you know, source, receive, manage and deliver sort of 400 million SKUs around the world. And we started down that path with like two day delivery, then one day delivery and same day delivery. And now, you know, we're working on 15 minutes. And so this sort of constant building of the expertise and this sort of weaving through, as Andy talks about in his shareholder letter, like not

Amazing. That's a lot of expertise and experience to be starting in there. So, Olivia, why don't you show us what it looks like?

Yes, thank you. I'm a supply chain planner and I was just handed one of our hardest challenges. I need to prepare for a new product launch with no sales history. The question is how many units, where, and when. Let me show you how I can tackle this with Amazon Connect Decisions. Traditionally, I would have to bounce between multiple systems, multiple spreadsheets, holding all the connections in my head. This is what most planners live in today. Let's see how Amazon Connect Decisions can help. I ask it to build a demand plan for the new products, a line of fitness trackers. Decisions finds comparable products in our catalog and does something traditional planning systems can't. It pulls in the sales team's own forecast for the new product line. It combines that bottoms-up sales intelligence with Amazon's forecasting models, built by the team that manages demand across 400 million SKUs. In the time it takes to get a coffee, I have a demand plan that previously would have taken me weeks. I can now make adjustments based on my experience from launching similar products. One of the things I've seen with other products is that demand can be different in the first weeks immediately following the launch, especially when there are customer incentives. Let me make that update. At this point, Decisions detects that there's been a manual edit and asks me for an explanation to better understand the reason behind the change. With a proper understanding of the change, it asks if I'd like to see the changes applied to the other SKUs in the plan. So with a single click, I can update the plan and publish it. And with that, the demand plan is already flowing into supply decisions downstream. Let's jump forward a few weeks. The trackers have launched and the demand plan is in motion. In traditional supply chains, this can mean database systems generating thousands of exceptions and alerts every day. Planners spend more time triaging noise than solving real problems. Figuring out which alerts actually matter is a full-time job. With Amazon Connect Decisions, it's different. Its agents have been monitoring inventory, supplier performance, and actual demand against forecasts. Instead of thousands of alerts, I see three prioritized exceptions. The most critical: a key supplier is trending four days behind, which will cause a stockout at two distribution centers within ten days. Decisions has already traced the root cause and flagged two other suppliers showing early signs of the same pattern. It has surfaced insight that would take analysts days to find before it becomes a crisis. Amazon Connect Decisions presents two options with clear trade-offs: one optimized for speed, one for cost. I pick the cost option even though the system flags the safety stock warning. And just like with the demand plan, Decisions asks why. I share that Atlanta is going down for maintenance so that inventory would be sitting idle anyway. The system successfully creates the transfer order and recommends I create a new rule to resolve similar issues automatically in the future. Decisions captures what your best people know, applies Amazon's 30 years of agile decisioning experience, and helps the entire team make better decisions every day. Plan, decide, act, learn. That's Amazon Connect Decisions.

Thanks, Olivia. Now, watching that, I have to say, you saw how the agent kind of showed up in different ways. It felt very seamless and fluid, and I want it like that goes back to the new design experience. It's interesting now that I saw it, kind of the thoughts on that.

Yeah, about. All right, now we're going to go from decision making. I want to go to our next lesson, which is Amazon Connect Talent. Now, this solution is focused on the area of high volume hiring, which we'll spend more time on. But start with what experience does Amazon have to bring to this domain?

That's 250,000 people hired. That's a lot. But I want to start with the hiring manager experience. So how did you think about that?

All right. We have Olivia come back and show it in a moment. But I want to go then to from the hiring experience to the candidate experience and kind of what was your mental model around creating that?

Got it. All right. Well, Olivia, why don't you show us what it looks like?

Thank you. I'm a recruiter for a large retailer, and I'm hiring for seasonal retail sales associates at multiple locations. The job description is what you would expect: responsibilities, qualifications, what we offer. It's pretty standard. I've loaded the job into Talent. So let's see how we get started. The science behind the application is impressive. We've directed agents to review the job description, identify competencies, create interview questions, and build an assessment rubric for this role. Talent has completed a review and generated a full plan. I can inspect every part, making changes and tweaks to ensure it fits my needs. Now that we've got this set up, I can get it posted and start interviewing. Talent lets candidates interview on their own timeline in their own environment. No scheduling bottlenecks, no pressure to perform in a narrow window. Candidates get a consistent experience regardless of when or where they interview. And because there's no scheduling friction, we're not losing great candidates who can't make a Tuesday at 2 p.m. So let's see what the candidate experience looks like. Hi there. Thanks for joining me today. I'll be conducting a phone screen interview for the seasonal retail sales associate position. Let me start with my first question. Tell me about a time when you helped a customer who had difficulty explaining what they were looking for. How did you figure out what they needed? Sure, yeah. There was a time when a customer came in looking for a gift, but they really had no idea what they wanted to get. They seemed pretty overwhelmed, so I actually helped them find the right gift. That's helpful. Sounds like you took the time to work through it with them. Can you walk me through what you actually did to figure out what gift would work for them? Yeah, I started by asking a few simple questions like... So what used to take weeks of phone screens can now be done in days. And every single interview is structured, consistent, and aligned to the evaluation plan. I'm not spending my time on logistics anymore. I'm spending it on the decision that actually matters: who to hire. So it's been a couple of days in, interviews are done. We've completed 80 interviews in days versus what would have taken me weeks previously. Let's assess the candidates. Amazon Connect Talent has assessed each candidate against the evaluation plan we created earlier. It gives me standardized data that I can use to make a decision. It's prepped the data for me, but it's not deciding who to hire. That's up to me. Talent anonymizes candidates, helping to reduce human preconceptions that come with seeing candidate demographics and personally identifiable information. For this first candidate, I can see how they performed against the competencies. We defined their top strengths and growth areas. I'm evaluating them based on job-related performance and not gut feel or personal affinity. I can also get a detailed view by going right into the interview transcripts to see how Talent arrived at its scoring and where the gaps were. In the end, the decision is mine, one I can now make with standardized data. Talent takes Amazon's decades of hiring expertise and puts it to work for every company. It removes the toil and gives back to recruiters and candidates what matters most: time, consistency, and the confidence that a qualified candidate got the job.

Thanks, Olivia. Just come back to what you said. I feel like you see that like the change in the candidate experience, like the flexibility they have, but also just getting that like clean, fair shot, right? It's really a game changer and how that works. Excited to see how that's going to affect hiring. All right. Now we're going to move on to our final area, which is Amazon Connect Health. And I'm going to just start with off the top. Like why is Amazon qualified to be in this area?

Makes sense. All right. Now with Amazon Connect Health, you're looking across both the patient, but also the provider experience, which is a little bit of a different take on it. So interested to hear about that.

Yeah, right. It's staggering when you think about what we could unlock if we had a better approach. So let's go actually run a video of providers talking about what this experience could mean for them. So why don't we take a look at the video?

Sure. The following people you are about to hear from are real working medical professionals compensated for the time spent in this discussion. We didn't script a word. We just wanted their honest opinions. We asked them about challenges in patient care.

Today, instead of having a face-to-face conversation, I'm sitting there typing the entire time, especially at a devastating time. It's extremely hard for patients to absorb the information that they're given.

I didn't sign up to be an administrator. I didn't sign up to be a secretary. And sometimes you feel like that's what you're doing.

Having this is very exciting. I like how it's streamlined versus just who's checked in. To have a brief summary would make me much more confident walking into the room. I love the fact that you have the update since the last visits with the lab results.

You know, I'm making direct eye contact and just talking naturally, you know, conversationally with the patients instead of, you know, looking up and down from my computer. Reminders of what to do is very helpful for primary care specialties. The suggested codes, that's amazing. That kind of interconnection, that's huge.

Post-visit summary, really useful. It allowed me to put together a more complete and accurate representation of the patient visit and notes. A lot of times you get these after-visit summaries and they're still as if they're talking to me. This I think a patient can read and see and understand.

The asynchronous text to remind to schedule follow-up, I can come back and do this when it works for me, is really helpful. More time directly interacting with the patient rather than trying to record data, you're able to interact with your patients and build up that personal relationship.

I'm Matthew Arneiter, I lead up innovations at Netsmart, and Netsmart is a health tech company that provides solutions to the behavioral health and post-acute communities. The relationship we have with the Amazon Connect Health team really allowed us to innovate quickly. The ultimate measure of whether these technologies are successful is whether it allows healthcare to be more human. If we had a way to alleviate the documentation burden, I think every single person that is a physician will cheer.

Awesome. Now, on the topic of healthcare, obviously sensitive kind of privacy, how as a provider, like trust an AI system to help them like this?

So certainly an incredible set of new solutions around the Amazon Connect family from Decisions, Talent, Health, of course, joining our customer solution as well. So any final words you'd like to leave folks with?

Amazing. Thank you so much, Colleen. Thanks for your insights and for the new announcements. Thank you very much. Take care. All right. Now, so far this morning, we have talked a lot about transformative capabilities that agents allow in solutions. And while many of those will come from us, like Amazon Q and Amazon Connect and the new family, we also know that the vast majority will come from all of you and being built by all of you. So to enable this, we've teamed up with OpenAI in a big multi-year partnership spanning models, infrastructure, as well as other solutions. So to tell us more off the top here, we're going to hear from OpenAI CEO Sam Altman.

Hi, everyone. I wish I could be there with you in person today. My schedule got taken away from me today. I wanted to send a short message though, because we're really excited about our partnership with AWS and what it means for our customers. And I wanted to say thank you to Matt and the whole AWS team. AWS is one of the companies that changed what builders could do. Cloud made it possible for startups, developers, and enterprises to spend way less time on infrastructure and much more time building products, serving customers, and bringing ambitious ideas to life. This was a huge platform shift, and AWS really defined it. I really lived through it during my early career at YC. Now, I believe that AI is going to be an equally important shift. We're still early, but the pace of progress is remarkable. Models are getting more capable every month, and it's not slowing down. AI will make it way easier to solve problems that used to require very large teams and very specialized systems. We're going to see AI lower the barrier to creating new products, doing important scientific work, bringing new ideas into the world, and much more. But to make that real for companies and particularly large enterprises, great models are only one part of the answer. These systems need to run reliably and robustly. They need to be secure, they need to scale, and they need to fit into environments where companies already run their businesses, and they need infrastructure that customers already trust for their most important workloads. That's what makes this partnership with AWS so important. We are co-developing an agent platform from the ground up, deeply integrated with AWS services and powered by OpenAI's most advanced models and tools, so that customers can build and run powerful agents in their own environment without worrying about the underlying plumbing. The developer platform for agents in the cloud, built to help enterprises move from experimenting with agents to deploying them at global scale. That's what we're trying to build today. Announced today, we're already seeing the major impact agents can make with products like Codex. Codex has changed from helping developers write code to helping teams automate more of the software development lifecycle and take on complex, multi-step work. That is why we are so excited to bring Codex to customers on AWS, making it easier for developers and enterprises to deploy and use Codex in the AWS environments they already trust. We're also excited about the infrastructure side of this partnership, including AWS training capacity, which will help us scale advanced workloads and make intelligence more broadly available. The long-term path to making AI useful for everyone is to build much more infrastructure, make models much more capable, and make these systems much easier for customers to use. And we are going to try to do all of those things. AWS made cloud the platform that a generation of builders rely on. This was amazing to watch. AI is now becoming the next great platform for this kind of building, and together, OpenAI and AWS are creating the enterprise platform for the future. The opportunity ahead of us is enormous, and the most exciting part is that this is not something in the future. It is starting right now. So thank you to Matt, the AWS team, and everyone here today. We are grateful for this partnership and we're excited to build this future together. Thank you.

Awesome. All right. Welcome back, Matt. Good to see you again and welcome Denise Dresser, the Chief Revenue Officer of OpenAI. And Anthony Liguori, who is our Distinguished Engineer at AWS Infrastructure and AI. Great to have all of you here. Thank you. Matt, I want to start with you. Just heard from Sam about the partnership. I want to hear from you. Why is this a big milestone both for enterprises and for AI?

Perfect. Denise, what makes you excited from this partnership?

Yeah. Well, so I've been at OpenAI for months now, and I've met with hundreds of enterprise customers. And what I hear right now is they're no longer in the mindset of experimentation and pilots. They really want to go full enterprise-wide, and they understand that to do that, they need to have powerful models. But even more importantly, they want those models in a trusted environment that they know and trusted infrastructure. And so for me, that's what's so special about this partnership. I think we have a joint mission around customer success and outcomes, but more importantly, bringing the power of GPT-5, our APIs, Codex, all on Amazon Bedrock allows our customers to focus on what matters. We talked a lot earlier today about building agents. That's a new motion for a lot of companies. So giving them the capability to be able to focus on that and to build and to innovate, especially right now when the technology is moving so fast, it is so hard for enterprises to keep up. So making it easier for them to do in a way that's trusted, I think is really an important element for our customers.

I agree, I think it's going to be a game changer. I know, Anthony, I know you're leading the technology part of the partnership. So how's it going from your perspective? It's been what, all of eight weeks?

Yeah. So it's been absolutely amazing. It's been wonderful working with the OpenAI team. We've had a lot of fun together. It's really just wild thinking about the fact that we had nothing eight weeks ago. And now we're here on stage talking about all that we've built together with a brand new partnership. Really, I think it's been a couple of things. One, today you can develop so quickly with tools like Codex. It's just really remarkable what you can build and how quickly you can build. And then two, both teams have been super well aligned in terms of the mission and what we're trying to achieve. We haven't slept a lot. That's part of it too, unfortunately, but that's because we're both so motivated to get these models into AWS and to really unlock this next generation of tooling for customers.

Yeah. Incredible speed. Just watching it from the outside, incredible speed and hopefully some sleep at some point. But maybe for the future. Yeah, exactly. But speaking of moving fast, in just the last 24 hours, things have also changed a lot. And we'll now be bringing OpenAI's leading models to AWS running on Amazon Bedrock. This will be in limited preview today already and GA in weeks now. Denise. Matt, what is now possible for customers now that we have this preview soon to be GA with this update?

Yeah, I think that's exactly right. And so starting today, 5.4 will be available inside of Bedrock in limited preview. And then within the next couple of weeks, we'll have 5.5 and all of the frontier models from OpenAI going forward. And like Denise said, this is what our customers have been asking us for for a really long time. Their production applications run in AWS. Their data is in AWS. They trust the security of AWS, and we've forced them for the last couple of years to have to get the great OpenAI models to go to other places. And they didn't like that. And now I think we don't force people to have to make that choice. And kind of building your application where your models and your data and everything kind of lives in the same place inside of your secure environment is an incredibly compelling proposition for customers. I think it's one of the things that I think drives innovation. I don't have to think about how do I cobble these things together? They just all work together. It can be integrated into your permissioning system. It can be integrated into all of your applications. It's something that customers have been looking forward to for a really long time. It's one less thing for them to think about and contemplate in a world where innovation is happening so fast and companies are trying to keep up, keeping it simple and clear and pulling it together.

Now, certainly with models on Bedrock, we know that customers are building agents today. So in addition to having model choice on Bedrock, we also have Bedrock AgentCore. That helps customers take those models and then create agents that they can deploy and scale. But we're of course not stopping there. So today we're also announcing the preview of Amazon Bedrock Managed Agents powered by OpenAI, formally announced as the stateful runtime environment in the original state for. Remember that now, Anthony, you have been not sleeping and working on this for the past eight weeks. So I want you to share like what is exactly Bedrock Managed Agents and how does it work?

Sure. So Bedrock Managed Agents allows customers to take OpenAI's frontier harness and models, and combine them together in an agent that lives natively within AWS. It really consists of three parts: the runtime, which is how you configure things like skills, memory policies, and the tools that the agent can access; the environment, which is where the agent lives. And so it's an EC2 instance, a Fargate task, or any other compute that's available within AWS. This is really how the agent integrates deeply within AWS, being able to access resources that are only available behind VPC using IAM policies so that you get CloudTrail and IAM-based organizational governance. And then finally, the inference API, which is the way that you actually interact with the agent, how you talk to it. You can use existing software, existing web applications, or you can use the OpenAI SDKs to integrate the agent deeply within your application.

Got it. Now I want to go back to where you started around the harness, because that's such an important part. Talk to me a little bit about like the innovation happening with the harness aspect of it.

Absolutely. Models can generally figure out how to use tools in lots of different scenarios. But when the harness is trained specifically with the model, you get something that's special. I think the best analogy is if you think about a football team, a football team is going to constantly practice the same plays over and over again. So when the time comes, you're able to really do well executing those plays. Ultimately, the harness is the playbook for the model. And so by training the model with the harness, you ensure that you're going to execute in the best way the model can.

Got it. So that tight connection is super important. And the outcomes, the customers. Absolutely. Now you also, I think I was following it, kind of sounds like affectionately three steps for customers and Denise's point, keeping it simple is that like as a customer using it, what does it look like?

Yeah, it's actually remarkably simple. If you have an application that's performing inference today, the only thing that you need to do is call the runtime API to determine how you want your agent to act. A really simple agent can just have a very straightforward memory policy, or you can add a bunch of skills if you want to really customize that agent. Then you create the runtime, you decide where you want the agent to live, an EC2 instance, or you can even use your laptop if that's what you want to do, but then that's it. Your application will work the same way it's always worked in terms of interacting with the inference APIs, but now you have inference that is stateful, that is able to keep memory and history of what's happened before, and most importantly, can access the entirety of AWS and can be embedded deeply within your applications.

Amazing. All right, so another part of our announcement in our partnership is we're also bringing Codex on AWS, which we're also announcing as preview today. Denise, I would love you to talk more about this.

Yeah. Well, first of all, Codex is our agentic software development lifecycle. It is absolutely on fire right now. We grew from 3 million weekly active users to 4 million in 2 weeks. And I'm sure it's going to keep growing, but it's beyond just coding. It is truly agentic work here. And what I think is really important is for all knowledge work, but the power of being on Amazon Bedrock brings another level of reliability, scale, infrastructure, and trust to this world, which allows when companies are thinking about identifying coding and innovation and work, they want to move at speed and scale in a trusted environment. I think that's what this partnership unlocks specifically for Codex with the reliability, which allows companies to move fast.

Amazing. All right. So that's a lot of announcements already, but it's still early. And so I want to hear from each of you. Where are we going next? Like from this partnership in the first eight weeks to when you look around the corner, what gets you excited? Maybe starting with you?

Yeah. Denise, how about you?

Well, for me, I think what has stood out in meeting the customers is the willingness to think completely different. I mean, CEOs especially are thinking about how do I need to reimagine my business? How do I need to transform everything that I do? That's a big proposition. But when you bring the power of what we're doing together, we make it so much easier for companies to move fast to actually achieve that. And when we take care of all the when, when on Amazon Bedrock taking care of all this amazing infrastructure, the best models, it frees up time for companies to actually innovate themselves and think about the bigger opportunity for them. And I think that's exciting. And for us to be able to help them do that is a really noble cause. And we're excited about it.

Yeah, I think so many enterprise leaders are hungry for that. Yeah. So Anthony, how about you? Maybe some sleep, but back on theme, probably not.

So in the last year, my life as a software engineer has changed dramatically. I am not 10 or 20% more productive. I am 10 or 20 times more productive. And so is my team. And seeing what that has done is just incredible. I never would have guessed this is the world that we would live in, and what's most exciting to me are new APIs, like Bedrock Managed Agents, where we can take that same proven harness and ability and start applying it into enterprise applications in their greater infrastructure. I think we're at really early days there, and I think like the next year is going to completely revolutionize the industry, right? You can imagine like what last year looked like, what the next year looks like, right? Unbelievable.

Only believe it. All right, just a few final takeaways, Matt.

Perfect. Denise, any final takeaways?

Listen, I think Matt captured it. But I think this is a partnership grounded in customer outcomes. And you can see the intensity of the work that we are doing and the speed in which we're doing about it, because I think we all believe deeply that this is an incredible opportunity, and it is what we're excited about, to see companies being able to bring these technologies together and just build. And I think, Matt, you said it best. It's like, imagine what we're going to see when we innovate together, but also when we're enabling our customers to build and what that's going to look like in the future. That's really exciting.

Yes, absolutely. Well, thank you so much for joining and being part of this event. We are very excited, as you can hear from all of us, to see how you will use agents to transform your business. Thank you so much.