AWS re:Invent 2025 | Colleen Aubrey, AWS

Clips
More from AWS re:Invent 2025

Colleen Aubrey

SVP of Applied AI Solutions (S-team)

AWS

play_circle_outline Colleen Aubrey discusses transformative implications of AI teammates for software development.

play_circle_outline AI teammates can evolve over time, improving collaboration and efficiency in teams.

play_circle_outline Organizations face challenges in adopting AI due to compliance and security concerns.

play_circle_outline Ensuring Safe AI Integration: Prioritizing Data Quality and Boundaries for Successful Enterprise Adoption

play_circle_outline Companies must navigate personal and professional data use with AI cautiously.

Info
Transcript

Colleen Aubrey, AWS

Colleen Aubrey

SVP of Applied AI Solutions (S-team) AWS

In this broadcast from AWS re:Invent, Colleen Aubrey, senior vice president of Applied AI Solutions at AWS, joins theCUBE’s John Furrier to discuss the paradigm shift from generic chatbots to sophisticated "AI teammates." Aubrey breaks down the launch of frontier agents – specifically for developers, DevOps and security – that are designed to collaborate, reason and improve over time rather than just performing isolated tasks. The discussion highlights how AWS is moving beyond simply abstracting infrastructure to abstracting work itself, allowing enterprises ... Read more

explore Keep Exploring

What is the concept of AI teammates and how does it impact roles and work processes in various industries? add

What is the speaker's perspective on the concept of a teammate in the context of AI development? add

What are the challenges and considerations for enterprise adoption of AI in contracts and compliance? add

What considerations should organizations take into account when developing effective workspaces for their agents? add

What considerations arise when blending work and personal interactions using a system like Alexa+, and how might customer preferences and privacy concerns impact this experience? add

bolt Powered by CUBE AI

Colleen Aubrey, AWS

search

John Furrier

>> Welcome back everyone to theCUBE's live coverage of AWS re:Invent 2025. I'm John Furrier, host of theCUBE. We're here for three days of wall-to-wall coverage, full day tomorrow as well. All the action, our 13th year covering AWS re:Invent, we've seen the evolution of the genius of cloud extracting away servers and hardware and systems and infrastructure to now abstracting away data and work as agents bring that whole nother step function level of change. Big part of the announcements here. Colleen Aubrey here is SVP of Applied AI solutions at AWS, formerly with Amazon, building out some of their largest businesses on the Amazon side. Great to see you again.

Colleen Aubrey

>> Thank you for having me job.

John Furrier

>> Thanks for making the time. So you're starting to see this next level. And we chatted about it last time we were talking on theCUBE in Seattle around Connect, how Amazon has all these kind of things that you can see. Now you see Swami and Matt and the team saying, "Wow, what if we think differently around software development?" And that's kind of the core. I mean, agents are like a side effect of that. You got frontier agents and you got Nova Forge, two big announcements that jump off the page that's going to enable that. But at the end of the day, it's software development transformation. And they dropped a legacy system, they blew up something on Monday. I saw that spectacle, but whether it's writing code for apps or migration, you're starting to see the new paradigm. How do you talk about that? I know you've been giving a lot of speeches. I heard you got great reviews of the analyst session.

Colleen Aubrey

>> Yeah. So I think certainly last time we talked, which was like in the summer, middle of the summer sometime, I think I had started to talk about this idea of AI teammates. And certainly, this has become more and more crisp for me. And I think our frontier agents with Kiro developer agent, the security and DevOps are examples of now I have an AI teammate. And as a developer, I'm managing this team, I'm delegating, supervising, I'm auditing, I'm giving feedback, I'm coaching, and then asynchronously work is starting to happen on my behalf. And so for me, this idea of AI teammates, I think it applies to every function, every industry, every business. And so that's a mission that we're on.

John Furrier

>> And you think about the frontier models and now frontier agents, the word frontier actually means something. And before LLMs had frontier status, which is either some people define as taking new ground or doing new things or capabilities. It was kind of boring, a chatbot. And then you got reasoning that became frontier models. Frontier agents have a similar thing where pre-frontier, they were just agents, but now they have more longstanding thinking. They can do tasks. And so I get that, but I want to get your thoughts on some of the industry jargon. So it's been said in the industry, AI is like having a bunch of interns and okay, they technically are teammates, but I don't want just interns. I want to have scope domain intelligence, Kiro, autonomous coding, that's some domain experience. DevOps agents, that's like having an engineer on call and then security. Those are the three agents came out. So that's the distinction between the collection of interns versus real teammates. How do you think about what teammates do? Obviously we all want to have the best teammates, A player, hire A players. So how does that apply to the AI side?

Colleen Aubrey

>> Yeah. So for me, a teammate is... It's a capability that's going to develop over time. So firstly, I agree with you, like having like a jack of all trades, something that does a little bit of everything, there's some usefulness to that. And I think there's a little bit... I might describe we've been in the jack of all trades AI era for a little while. And it's been... You get AI doing tasks. Can you just like do that one thing? Yeah. I can take care of that. That's awesome. And so, but I don't... That's not like an expertise. It's not like having an expert that you can work with that brings deep understanding of a particular area that really understands how that work needs to happen, that understands what good looks like. So I think with our frontier agents in the development space, it's an era we're very familiar with in AWS and at Amazon. And so we have a good amount of knowledge that helps us to develop now something that a teammate that's going to grow, develop, earn my trust, start to provide feedback, start to collaborate with me, will over time become better and better every day. And I think that's the thing. I want an AI teammate that gets better every day.

John Furrier

>> Matt Garman told me in my interview, his quote was, "Generic tokens are useless unless they know your business." And referring to tokens and tokens per watt, that kind of thing. And he's kind of getting at the whole generic reasoning or jack of all trades comment. And you look at Kiro, one of the things I liked about the news was it adapts to the preferences of the team and every team has their own nuances, right? Like, Hey, we like our code a certain way, or here's how we document or here's some of the practices we do. It might be different from team to team. They have their own language. So that's where it starts to get interesting to me. So how do you frame that? Because that begins to take into that 10x engineer mindset where, Hey, I'm one person, but now I have a team of agents that are going to actually do the work and they can go off and do longer form because they're frontier models. How do you think about that? Because that changes the domain intelligence. Now I have augmentation as the human in the loop in this case.

Colleen Aubrey

>> Yeah. And so for me in my area, I guess Connect is the furthest along this space where we're really trying to push in the direction of agentic experiences for end users, for people that are working in the business and for people that are managing the business to be able to work on a much bigger surface area. To be able to look at not optimizing tasks or steps, but optimizing whole customer experiences. And so for me, the agentic teammate starts to increase the surface area I can explore, but then also needs to be part of the team. Like you were talking about your daughter before, she's an Amazonian. Her language has started to change, the way that she talks about things starts to change. And I think that's what an AI teammate needs to do. Now you might start with, and certainly in the case of customer service, you start with the new knowledge basis, you have good SOPs that are well documented. I think that we'll increasingly see controls about the personality of the AI that's working in your team and we'll also see that learning start to filtrate through. But I think like the same as onboarding any person into your group, they come with a set of skills, they come with a set of working styles and ways of interacting and they learn and evolve over time.

John Furrier

>> It's work that learns you. I mean, it figures out your preferences. Matt also said... Stats on productivity said, 80 to 90% of the value will come from agents in the enterprise, an area that's been slow to adopt because of many reasons. One is compliance, contracts. No one wants AI in these contracts. There's all kinds of resilience bars and security. So I think this would... To me, opens up a path because with Forge, Nova Forge and agents, you essentially can go in and knock down some use cases to go down the road that you built with the team with Connect where how do I get my AI infrastructure, AI native foundation. Versus boiling over the ocean and trying to figure it out. So how do you see the enterprise adopting this? Because with this end to end workflow and coding, which is certainly being adopted, the enterprise now have this quick path to knock down some wins. I mean, Matt even was talking about ROI and the keynote. Usually don't hear ROI and keynotes at these technical conferences. But I think he's pointing out that you can knock down some wins to stay in the game, give the six months and they showed a curve where it goes up. You start to see the real value. How do you see that value for enterprise? How should they think about it as they try to figure out how to make these agents work?

Colleen Aubrey

>> Yeah. So I think what we've seen certainly within AWS, we're increasingly expanding the building blocks that I think are necessary for an organization to create the space. A space that they feel is safe and secure, where you can really then start to unleash the creativity of the people within that organization. And certainly, I think the development community is eager to do that, they're well positioned to do that. And I think that the technology is well-developed for that. If I think about all of the other functions in an organization, defining that space, and you know what it's like. I think we talk about like data and you don't want garbage and you need good quality, but you need good boundaries around that. I think organizations will increasingly find their path to lay this foundation where they decrease the segmentation of their data. They find ways of putting the parameters around helping for this to be secure. I think we increasingly see the capabilities to say, well, if AI is going to be part of your team, what are the permissions that they need? How do you think about their permissions? How do you think about observability, which is super important. Can I interrogate what's happening? Can I understand the reasoning? All sort of part of the interactions you would have with another person in your group. And I think once you create that environment, I think organizations and enterprise will be able to embrace that chaos a little further.

John Furrier

>> Yeah. You have a unique history, and we talked about this last time coming into AWS building the advertising business at Amazon. That's large scale. It was big numbers. You had to take a holistic view. We see Connect on that trajectory. I can almost imagine with Agentic that there's going to be an applied AI system to your title, Applied AI Solutions. Well, first of all, what does that mean, applied AI solutions? But my guess is there's a holistic view here with the scale of AWS and these three agents, just the beginning, it's act one. Will there be more Amazon applied solutions or how do you see the ecosystem playing in there? Because one of the things that's jumping out on these large scale like AI solutions is that co-designing things together has been a really big part of successful companies, whether it's Amazon, web services or Nvidia, the co-designing, the supply chain and the ecosystem, it's different kind of behavior because there's so much involved. How do you see the holistic Amazon Web Services picture?

Colleen Aubrey

>> Yeah. So firstly, for me, I guess the applied AI is really trying to hone in on getting AI to work in your business, in the operations, with your teams every day. And specifically, I'm focused on the business user and the business functions. And so that's why you've used this term, applied AI is getting it to work. And I think that last mile is hard. That last mile of getting AI to meet your regulatory requirements while still getting the benefit of what AI is, which is the reasoning and the generative power. If you just want predictable outcomes, you should just use ML. And so there is this, I think, work to do and the blending of the software, the ML, the AI, and the environment that that works in to actually get that reliable business operation.

John Furrier

>> We actually had a guest earlier, Ana Pinczuk from SentinelOne, and they're all on Amazon Web Services now, but they're co-designing at the security level with all the Amazon scale. So they're taking advantage of those things. So that seems to be the pattern, a feature, not a bug, where it's like, okay, you can choose to do your own thing or come in and apply this. Now she's a senior leader. Obviously she's a CTOC product officer, but she's betting her business on AWS because of that scale.

Colleen Aubrey

>> Yeah. And I think for me, one of the, I guess the opportunities that we have, and you mentioned, I've worked in the ads business before, is that actually I have this very large diverse business within Amazon. Where we are putting AI to work and have an opportunity to learn and refine that and actually work with those teams to develop something that we think actually this can work at scale, it can work reliably in a business and it can work with great impact and ROI. And now we can productize that and put that in the hands of customers.

John Furrier

>> And you're focused now on this large scale business problem is operational-

Colleen Aubrey

>> Yeah, so-...

John Furrier

>> architecture system approach?

Colleen Aubrey

>> Yeah. I think Connect gives me a nice starting place. From a platform point of view, in terms of like, this is the software and the data layer, it's about managing high scale workflows with a lot of variability, with complex business requirements. And this is not something that is specific to contact centers. This is actually the capabilities you need in many functions. And then on top of that, you have these expert AI teammates, and I think we'll increasingly develop them in the customer service case, but we're exploring a number of new functional areas and vertical areas where we can bring together these AI teammates on top of this quite flexible platform and then put something in the hands of business-

John Furrier

>> The vertical apps of vertical agents are going to be really valuable, we think. Another announcement I want to get your thoughts on is the AI factories. You mentioned systems. That is not outpost. It's not a bunch of edge boxes. It's a whole different approach. And then Mobile Congress is coming up, will be at NRF retail event. You're talking about an edge now coming around the corner where you have not just businesses wanting to have an on-prem AI factory with sovereignty. You have a hyper-converged edge environment where spectrum and wireless and ethernet, wifi collapse together. You built a little AI factory in retail out like Nordstroms and you walk in and I might want to have my preferences identified. Now just walk out, they go privacy. They got all the cameras, but I might want to declare, "Hey, recognize me and go get my model and see what's on my cart on my PC at home."

Colleen Aubrey

>> Yeah. I mean, it's interesting, like with Just Walk Out, I think the team have done a very nice job here because it really is the unidentified movements. There's no individual identification. It is anonymized short-term ID associated with the movements of a person in the store, no faces, no names, no anything of that-

John Furrier

>> Full privacy.

Colleen Aubrey

>> Full privacy, but also So the power of full visual reasoning. And so I think this is going to be an interesting area and that whole area of visual reasoning, we're exploring a bunch of different applications of where would you put visual reasoning to work in an organization that solves other adjacent problems. And we're early in exploring them. And again, really doing a lot of experimentation within Amazon. It gives me a very nice canvas to play in.

John Furrier

>> Well, we're going to have this conversation in the World Conference for sure because we're going to table the conversation of if I'm an employee of a company and I go to retail outlet on my personal time with my business phone, I might want to be identified. And then an agent can give me a VPN. So there's a lot of things that I see around the corner here. Maybe I'm just riffing and dreaming, but when you have that kind of system scale, new scenarios pop out.

Colleen Aubrey

>> There's a lot to unpack here. Panos and I, Panos leads devices team at Amazon Alexa+ team. So Panos and I have this conversation of like, "How do you think about the blending of my work and my personal?" And let's say with Alexa+, will people want to do both with Alexa+? Will they want to interact on an echo show on both dimensions? And Panos and I go back and forth debating like what will people's preferences be? Where will customers go with this? And then how do we think about the privacy, the data, where does it start and end? How do you keep an experience really smooth, but also deal with many of these questions that you're raising? I don't know what the answer is.

John Furrier

>> Hypothetical questions, but it takes the conversation of the frontier agents to the next level because the act one of Kiro autonomous coding with DevOps and security is saying, "Okay, this is Amazon in a box basically. Here's your core team. Get basically engineers working for you for Amazon." For companies, that's great. Then they could also have other teammates saying, "Hey, figure out what an employee may want." So again, this kind of takes the... If agents continue to provide this work, new at scale use cases emerge that isn't just an ecosystem partner saying, "Hey, we're using agents for some domain specific applications, which could be great. I was talking to Elastic and Glean. They love it. They do in search." That's great. Check. But the bigger picture is... What I like about Connect is its AI native at scale that has all these other benefits.

Colleen Aubrey

>> Yeah. And I think earlier today in the AI keynote, one of the things that I said was like, I think what's exciting is I don't know if we've even imagined really how we will work, what we will work on, how this collaboration with AI teammates works. And then how does that work in my personal life versus my work life? And what's this... Is there a divide? Is there not a divide? I think there's many, many open questions. What's super exciting is certainly for me, I'm constantly surprised by how many companies are like a little more adventurous than I might have expected in the past and a little more willing to be disruptive. And for some organizations, that's the way that you operate every day. But for others, there are different styles and different approaches and they have different history that they're pulling with them. It's interesting to me that I feel like more and more organizations are feeling ready for the disruption.

John Furrier

>> That's why I like your business focus because I think from the boardroom to the dorm room, there's demand. Every boardroom's like, "We need to have a position in AI." And we were speculating on the a couple of weeks ago that there's no fast follower action here because in other markets, you'd be the fast follower, you kind of wait and see what happens. There's no time. If you're not in that, you could be an extinction event for a company. So I think the business people kind of see that. They just don't know how the mechanics work in the stack or all the security concerns that they go through. So I think there's definitely demand for that. And I think this re:Invent of the 13 years is a special moment because it's for the first time the ground shifting a little bit to abstract, like I said, abstract away data and work for more things. More things are going to happen.

Colleen Aubrey

>> And you know, for me, what's interesting is I think the one way door decision... Your daughter will recognize that. The one way door decision is to not get started. Integrating AI and AI teammates into your business to change how you deliver products and services to customers, there's no opt out on that. If you opt out, you're opting out of business. And I think that's the existential thing. So if there's no opt out, you have to opt in. And the one way door decision is not starting today. But I think once you get started, it is going to be a journey and it will be a constant sort of discovery and iteration and the technology will continue to change rapidly. So I think there's a fluidity that organizations need to embrace, but the really existential question for me is like, if you're not starting today, that's challenging.

John Furrier

>> I did like how Matt put the slide of three to six months in, the agents will behave part of your team. He actually shared that from internal data. So I think that's motivating, gives people a lot more confidence. Colleen, my final question for you is for the folks watching, you're inside the machine of AWS and Amazon, you're here at re:Invent, you have the keynote, you're talking to all the analysts. What's your big takeaway? If you had to boil it down to the key things that this re:Invent is delivering in terms of the 20-mile stair or a future path, how would you share your opinion on the relevance, the key things that are the most relevant?

Colleen Aubrey

>> So one thing I hear a lot of organizations talk about is automation and efficiency. And I get it. That seems interesting, but I think it undersells the opportunity ahead of us. This is about transformation. You have to be willing to put everything on the table. How work happens today, all your working assumptions, every bottleneck you've optimized around, every SOP, you have to put it all on the table because the way that you can work in future, the only thing I'm confident on is it's going to be dramatically different than today. And unless you're willing to do the transformation, I think you're going to only partially realize the benefits.

John Furrier

>> Awesome. And I've heard some people say, nail your determinants like workflows that you want to keep down and then let the AI help you grow that from there.

Colleen Aubrey

>> Yeah.

John Furrier

>> It's great stuff. Colleen, thank you for taking the time. I know you're super busy. Thanks for sharing with theCUBE. Appreciate your time.

Colleen Aubrey

>> Thank you for having me, John.

John Furrier

>> I'm John Furrier at theCUBE. Day two, coming to an end here. Again, really a moment for AWS as the data and the agentic era enters in. I'm John Furrier. Thanks for watching.