In this keynote analysis from AWS re:Invent 2025, theCUBE’s John Furrier joins
analysts Paul Nashawaty, Zeus Kerravala and Sarbjeet Johal to unpack how Amazon
is redefining cloud infrastructure through the lens of agentic AI. The panel
breaks down Matt Garman’s declaration that "agents are the new cloud," exploring
key announcements surrounding the Nova model family, AgentCore and Amazon
Bedrock. The discussion highlights AWS’ strategic pivot from merely abstracting
infrastructure complexity to abstracting work itself, effectively bridging the
gap between professional coders and "citizen developers" while unifying the
experience for builders at every level. The conversation digs deeper into the
practical realities of enterprise AI adoption, emphasizing the critical role of
security, governance and compliance in moving from proof-of-concept to
production. Kerravala, Johal and Nashawaty analyze AWS’ vertically integrated
approach – spanning from custom silicon like Trainium and Inferentia to the
application layer – and how this full-stack strategy allows customers to train
models on proprietary data with improved price-performance. The group also
debates the evolving competitive landscape, noting how AWS is equipping
organizations to build autonomous, long-running agents that function as
teammates rather than just tools.
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
AWS Marketplace Newsroom . If you don’t think you received an email check your
spam folder.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open the link to automatically sign into the site.
Register for AWS Marketplace Newsroom
Please fill out the information below. You will receive an email with a verification link confirming your registration. Click the link to automatically sign into the site.
You’re almost there!
We just sent you a verification email. Please click the verification button in the email. Once your email address is verified, you will have full access to all event content for AWS Marketplace Newsroom .
I want my badge and interests to be visible to all attendees.
Checking this box will display your presense on the attendees list, view your profile and allow other attendees to contact you via 1-1 chat. Read the Privacy Policy. At any time, you can choose to disable this preference.
Select your Interests!
add
Upload your photo
Uploading..
OR
Connect via Twitter
Connect via Linkedin
EDIT PASSWORD
Share
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
AWS Marketplace Newsroom . If you don’t think you received an email check your
spam folder.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open the link to automatically sign into the site.
Sign in to gain access to AWS Marketplace Newsroom
Please sign in with LinkedIn to continue to AWS Marketplace Newsroom . Signing in with LinkedIn ensures a professional environment.
Are you sure you want to remove access rights for this user?
Details
Manage Access
email address
Community Invitation
Deepak Singh, AWS
In this keynote analysis from AWS re:Invent 2025, theCUBE’s John Furrier joins
analysts Paul Nashawaty, Zeus Kerravala and Sarbjeet Johal to unpack how Amazon
is redefining cloud infrastructure through the lens of agentic AI. The panel
breaks down Matt Garman’s declaration that "agents are the new cloud," exploring
key announcements surrounding the Nova model family, AgentCore and Amazon
Bedrock. The discussion highlights AWS’ strategic pivot from merely abstracting
infrastructure complexity to abstracting work itself, effectively bridging the
gap between professional coders and "citizen developers" while unifying the
experience for builders at every level. The conversation digs deeper into the
practical realities of enterprise AI adoption, emphasizing the critical role of
security, governance and compliance in moving from proof-of-concept to
production. Kerravala, Johal and Nashawaty analyze AWS’ vertically integrated
approach – spanning from custom silicon like Trainium and Inferentia to the
application layer – and how this full-stack strategy allows customers to train
models on proprietary data with improved price-performance. The group also
debates the evolving competitive landscape, noting how AWS is equipping
organizations to build autonomous, long-running agents that function as
teammates rather than just tools.
In this interview during theCUBE's coverage of AWS re:Invent, Deepak Singh, vice president of engineering at AWS, joins theCUBE’s John Furrier to discuss the rapid rise of agentic workflows and the philosophy behind "spec-driven development." Singh shares insights into the explosive adoption of Kiro, revealing that over 250,000 developers engaged with the tool within its first three months. The conversation unpacks how Kiro bridges the gap between the creativity of "vibe coding" and the rigor of engineering best practices, allowing developers to convert conve...Read more
exploreKeep Exploring
What event is taking place in Seattle, and what upcoming announcements are anticipated?add
What was the launch announcement and performance of Kiro at the AWS Summit in New York?add
What challenges do software developers face when writing and maintaining applications in a team environment?add
What company exemplified improved efficiency in project management by effectively specifying their work, and what were the results of these improvements?add
What is property-based testing and how does it enhance the testing process compared to traditional unit testing?add
>> Hello, I am John Furrier with theCUBE here in Seattle, Washington at the Amazon Web Services Headquarters at re:Invent Building, breaking down all the action with a agentic workflows, spec-driven development. Deepak Singh is back on theCUBE. Deepak, great to see you. Thanks for spending the time here in your home turf.
Deepak Singh
>> Yeah, it's great to see you again.
John Furrier
>> In the re:Invent building. We've got re:Invent coming up, and exciting announcements coming. I can feel the hallways talking. I can get a sense. Agent's going to be hot. We spoke in June, in the summer, kind of the halftime report we did on spec development coding, obviously with the code assistant, Kiro, news is out, generally available. This is one of the most exciting areas right now in the market. There are two hot trends. One is AI infrastructure, which everyone's talking about all the CapEx. That's the enablement, that's the physical hardware, the software and the clustered systems. That's enabling mass performance of software stacks. And this is kind of where you're leading. So give us an overview of Kiro right now. Where is it at? What's the status?
Deepak Singh
>> Yeah, I'll actually go back to our last conversation, which was at the AWS Summit in New York earlier this summer, which was when we announced Kiro to the world. And Kiro was launched at that time as an IDE, a development environment, for spectrum and development, for a way to do coding and build software, which was more robust, more maintainable, and with the core being this idea of specifications. And just right at the New York Summit in our first few days, in our first three days, we had over 100,000 developers use Kiro. And in just the first three months, we had over 250,000. So it was a great start. And people have been doing lots of meaningful work with it. Earlier this week-
John Furrier
>> What was the appetite? Why is it so popular? What was your take on that?
Deepak Singh
>> What really made excited people was people are driving agents to write software. There's many ways of doing it. Kiro does it really well too, where you are sitting inside a conversational interface, you are talking to an agent, and you're driving the code that it generates. The challenge with that is five days after you've written an application, you've kind of forgotten why. In a team environment, it becomes even more interesting because three months later, nobody has any idea why you wrote the software you did. And even for yourself, as you're evolving a piece of software, as you're starting with something that already exists, it becomes a little challenging to have the agent have the right behavior. So we started talking to our senior engineers and asked them, how do you write software? How do you build applications? And there's a tenet for our principal engineers whom you met, a few of them. And the tenet is called illuminate and clarify. And what illuminate and clarify basically means is senior engineers, our principal engineering community, is really good at helping people understand the problem, break it into smaller parts, and maybe work with others to clarify exactly what they need to do. And they often start by writing down a specification, maybe on a piece of paper or on a whiteboard. So we took that concept and converted it into what we now call spectrum and development. So in Kiro, you can start talking to an agent, to the Kiro agent, and say, talk about what you want to build, how you may want to build it. And at some point, you like where you are and you say, "Let's convert this into a specification." Or you can do that right from the beginning. And Kiro will convert that into a set of requirements. Once you're ready and you like those requirements, we use the stories and everything, you can convert into a design. Again, you're collaborating with this agent. You can take this design and show it to somebody else. And at the end, you convert it into a set of tasks which get executed by the agent. So this way of keeping the fun of AI-driven development, but bringing more structure to it is what excites people about Kiro.
John Furrier
>> I like this team approach, because there's always the old expression in tech, meet the developers where they are. And so talk about the environment that they're in, one. And two, when you talk about crossing code like across workflows and security, these all come up a lot. How do you see that playing out with Kiro? What was the response there? Because people got Slack channels, they got WhatsApp over here. We have a lot of ways that people communicate. And how does Kiro kind of embed all that?
Deepak Singh
>> In the end, they communicate through Git. Developers. Git is-
John Furrier
>> That's the system of record.
Deepak Singh
>> Yeah, that is very-
John Furrier
>> doesn't lie.
Deepak Singh
>> And actually the fun part with specs, specs are also stored in Git. In the end, they're just markdown files. And what we find is there's this digital identity and fraud prevention company called Socure. They were able to just, because they were able to write things down as specifications, they were able to take projects which were taking them weeks and convert them into a couple of days worth of work as they were evolving their code base. We've seen SmugMug and others have similar success with Spectrum and development because it allows you to take your mental model of what you want to accomplish, convert it into a more structured way of working, but it's structured both from a human perspective, but the structure also helps the agent give you better results. And that part of it is what I think resonates really well with our customers and why we have seen the interest in Kiro and beyond, as agentic capabilities continue to evolve, we'll keep doing that. I'll give another example. So we went generally available and we added this new capability called property-based testing. And property-based testing flows directly from the idea that you have a specification, but it combines a couple of concepts that you're very familiar with, AI and mathematical approaches. You've heard us talk about neuro-symbolic AI for example. But essentially what Kiro does is it takes your requirements, your spec, and extracts the things that you want that code to be able to do. And it creates all these tests that are testing the properties of your application, which goes way beyond unit testing. Which you can write a unit test, it may pass it or it may not. But here the agent is saying, "Here's all the things that code should functionally do. We are going to actually test if it actually does that." And so what it means is your code is much more likely to be actually good and do what you want it to do. And all of that flows from the spectrum and development paradigm that we are very-
John Furrier
>> It really is a mature product, I mean mature methodology for mature teams. So you mentioned vibe coding, people do vibe coding of course, but now you have deploying. Cloud-native stuff we've been talking about for over a decade now. People want to deploy in applications. So I guess my question with re:Invent coming up, you're seeing a lot of people in the DevOps community, especially cloud-native, folks I talked to, they see the AI world, I won't say , but there's a coming together of AI-native and cloud-native kind of doing the same thing. So how do you see the Kiro bringing the developers? Because cloud-native DevOps serves developers. They create the environment for deploying applications. So all those practices with Kubernetes and all the containers, all the things that we've been working on as an industry are going to feed right into a scalable agentic AI. What's your vision on that?
Deepak Singh
>> The way to think about it is how, you can say shift left. But even today, with Kiro for example, you have this idea of hooks. Kiro has this functionality called agent hooks where when you write some software. When you check in the software, you can automatically run that hook or run a hook against it. The hook could be something like make sure that it's secure, run a security scanner on it. Or make sure that the code is optimized for certain things. Or run it against a well-architected review, for example, and you get that feedback. That's great. Kiro allows you to do it. But increasingly with agents, you can actually move all of that work further left and say, "Make sure you're writing software that is already highly available, need security guidelines." I've actually seen people do that where, in their prompt that drives a specification, they'll say, "Write this application using industry best practices," and then your property-based testing will actually test for those best practices. Without you having to actually do it, it's automatically doing this. So the way you think about development change. It is your best practices now, the testing that you do almost becomes an audit rather than let's make sure this is actually the case. It's the code should already be there. That's your belt and suspenders. And how the industry and teams think around this is going to be fascinating to see.
John Furrier
>> It's interesting. And just to riff with you on this, because I think the cloud-native, we were at KubeCon from day one when Kubernetes hit, and so we've been following it as you know, it's funny, all the cloud-native engineers that did the DevOps movement from the early days to now, they did the hard work. They rolled up their sleeves, they built it by hand as an open source community. And when I talked to cloud-native alpha engineers and leaders, they're like, they're so mature now. They've matured up I feel like. When they look at the AI world, it's like, well, there's a lot of buzz there, but what does it mean for me? I'm well ahead on, we've been deploying infrastructure at scale, cloud-scale. And then the AI feels a little bit foreign and they want to look for where . I won't say they're there, but it's more of just a mismatch of the culture. And what I find is that there's so much affinity, but the cloud-native community's like, "Well, where's the specs? It's not as clean." And I think with AI coming into DevOps, it feels like a relief for cloud-native. Because what Kiro's pointing to is, hey, a lot of the stuff that you were doing that was the hard standing up all this capabilities and workflows, it's easier. What's your view? Because you're seeing both sides. You were part of the cloud-native movement and now you're also seeing the other side where AI's infusing in. It's not an AI app to say, "Hey, do all my work for me." There could be a little help there, but it's a step function, value change.
Deepak Singh
>> We see this in how our teams are working. If you talk to the Kiro team for example, they use Kiro to build Kiro. So we learn a lot from that. At Amazon, the company, we use Kiro. So Kiro has an IDE and now we have a CLI. The CLI has been around for a while as well, and we see how people write software evolving. I'll give you a couple of examples. So I think actually stepping back a little bit, we want developers to have the fun. Vibe coding is a lot of fun. You don't want to lose that. People get very excited, they write software, but we also want to make sure that the software's not a one and done. You're actually evolving it, you're working with other people. So you keep the fun of vibe coding, but bring in some of the best practices. And because AWS has a very large customer base that is building production applications, use network systems, there's a satellite company that is using spec-driven development, that allows us to very quickly learn how do you bridge the, I won't say call it a gap, bridge the connective tissue between the fun of vibe coding and the desire to build maintainable robust code. And of course within our company, where the software we write has to be maintainable, we have teams building production systems at scale almost completely using AI. These techniques start to evolve. A lot of these ideas that we have around spectrum development came from talking to customers, talking to senior engineers on how do you keep the fun while allowing you to write robust maintainable software that will last for years.
John Furrier
>> And I think I bring up as cloud native intersection with AI and AI native builders because, as customers, I talk to a lot of your customers, we're a customer, I would love to have an agent know my infrastructure so I can do more within Amazon and do more coding. Because cloud native has operation. DevOps is ops. There's dev and ops. So there's like you got to run operations and then you're still building new stuff. So this is where I think I see the thread with Kiro saying, hey, you're going to have the best of a software development environment and be more productive, but at the same time you can run it. So you're inside Amazon Web Services, so you have visibility into how to run, I won't say complex, but a lot of services. When do I use serverless, when do I use this database? There's a lot of choices. And so I'm thinking Kiro would be helpful here.
Deepak Singh
>> So in fact, if you look at it, we have this in the CLI right now, you can create what I call custom agents. You can actually write a custom agent that says, "I am a Kubernetes engineer or I'm a DevOps engineer, or I like writing software that way." It automatically rolls up a profile. It'll create a profile, it'll pick up the right MCP servers, it'll pick up the right steering files and it'll create almost an environment where now, from that point on, you're writing like... Somebody could write a set of steering files and create an MCP server that says, "Always write software this way." And the rest of the software that will be written will be written with that in mind. It's a custom agent. It's almost specialized in, you can call it cloud native for example.
John Furrier
>> Call my Amazon operator.
Deepak Singh
>> You could actually create a custom agent that says John's Amazon operator, AWS operator. And it'll have the right MCP servers, the right set of steering files, and we will then write software the way you like operating with all the best practices inside an AWS environment. The next second you could say I'm a front-end developer and it's picking up the right MCP servers, et cetera, for working with CSS files and with Figma designs, et cetera. And you can switch back and forth between these custom agents that are specialized at other tasks and their manifestations of what you like to do. Or let's say at a company like Amazon or at any other enterprise, your enterprise best practices can get manifested like that. So it's a different way of building. In some ways, it's very familiar. In some ways, it's unique. But what we find is that developers who learn how to... Developers are artisans, development is art. This is an art to it. And the best developers are the ones who learn how to harness these tools, just the way they've learned how to harness programming languages or various frameworks over the years.
John Furrier
>> And you brought up the teams piece, how has that changed? Because in the old school days you do a spec market development, here's the market, here's the specs, and you had stakeholders have input into that. Then it goes to the coders and the coders the old way. But how do you bring in that front-end, say UX designer or artist on the team? Can they be replicated? They have input? How do you see Kiro interfacing with stakeholders? Because I might want to have my UX person there.
Deepak Singh
>> Here's a fun part of it. We've had examples where a dev manager has built the first prototype and then handed it off to the engineers. It's like, "Here's the prototype I built using specs. Here's the spec for it. Here's the steering files. I like how it looks. What do you think?" They're like, "Yeah, it doesn't look bad." Then they go and take into production. We built an internal dashboard for our internal developers to look at their usage. The whole first version of it was built by a product manager. He wrote down the spec, built the prototype, with synthetic data, and then hand it off to the dev team and the data folks to complete it. We see our customers doing that all the time. A spec is a set of requirements that can be written by product manager and then hand it off in collaboration with the dev team. The key is that it doesn't have to be linear. It's not like I'm writing it. They all working on the same code base, its own system, and you don't have to be an expert developer because your agent is helping you understand and read the code base. You can start even on somebody who's not familiar with the code base or not a full-time developer can start. But in the end, Kiro is a-
John Furrier
>> It's a fast way to create innovation because you don't have to have a big team. The spark could come from an engineer on the weekend.
Deepak Singh
>> Anywhere, yeah.
John Furrier
>> It's raining this weekend. Instead of going at the park, I'm going to go write some code. Here's a prototype, kick it around.
Deepak Singh
>> Or in fact, right before we launched Kiro, we had to add a notification system to it, and it's a fairly complex design because it has to interact with a lot of systems. And the way it was built in a very short period of time was exactly this. One of the developers on the team over a weekend wrote the spec, built the prototype. In 24 hours, that verified this was a good working prototype. And another 24 hours, it had been merged into the main code. And then I think three days later, it was available in their launch in New York. So that's not uncommon.
John Furrier
>> Yeah, and this is a new way to do business with the coding. I love it. I guess my couple last final questions. One is a re:Invent teaser, but I'll come in a second. I want to get your learnings from Kiro now that we have the GA. Super high adoption. I could see why developers like it. Creates creativity, artisanship, the craft of coding and also creativity. But also you could put out product fast and then also learns the environment. What's the big learnings from the short time you've been accelerating this out of the market?
Deepak Singh
>> The biggest learning we've had from watching our developers, both inside Amazon and within our customer base use Kiro, the IDE and the CLI, is that developer, in the end, what you're doing is driving agentic behavior. You're helping the agent give you a good outcome. As developers learn how to harness these agents, whether it's through specifications, whether it's through better prompting, whether it's learning how to use MCP servers and writing the right steering files, developers who learn how to do that, learn that craft faster, are going to be super effective. We see teams working at pace at efficiencies that we've never seen before, writing really high-quality software in so much less time than they had before, whether it's in our customer base or whether it's inside the company. And it's fascinating to see. And we are learning a lot from that. And the design of the tools that we write, the way specs are presented to you, the way the CLI works are all driven by what we are learning from how developers are learning to use these tools.
John Furrier
>> It's the tools that connects to the brain that makes the output work better. And we're seeing that on the consumer side. I have friends that say, "I'm a prompt engineer now." They're bragging about how good they could use the AI tools. But that's a progression. They've been using it, hence they're getting better. That's to your point.
Deepak Singh
>> Yeah, there's this term called context engineering. Context engineering is only one part, to provide the agent the right context. Whatever we call it, it boils down to the people who learn how to harness these agents and work with them and embrace their capabilities and learn what their strengths are, et cetera, are the ones that are going to be successful. And with things like spectrum and development, they have toolkits that allow them to be more effective, write higher-quality code. But in the end, it is a developer, what's inside the developer's head, what they really want.
John Furrier
>> And they got to start somewhere. Like I was saying, consumers say, "Hey, what is this?" And then they learn the prompt. And now you're saying on the engineering side, do that. That's the same effect. That's the benefit of the systems.
Deepak Singh
>> Correct.
John Furrier
>> I guess my next question would be, and I'm curious, and I'm sure everyone else was too, how do I get started? I got to jump in and learn a new application. How does someone have the ease of use just to start using a chatbot, like ChatGPT. Like, "Hey, oh wow, that was cool." Is there a way to take baby steps in and get into that acceleration?
Deepak Singh
>> You said it earlier that it's great to meet people where they are. And with Kiro, both CLI and IDE, you get exactly that. The CLI is just an agent in your terminal, you're a shell, you type Kiro CLI, and you're off to the races. You're just talking to it. You can even ask Kiro, "What's the first thing I should do?" And it'll tell you what the first thing you should do with it.
John Furrier
>> I'm a newbie. Walk me through.
Deepak Singh
>> Yeah, "Kiro, I've never used an agent. Help me." It'll help you. The IDE is very familiar. It feels like any IDE that you used before. In fact, you can bring your own plugins that you may have already and bring it. You can import them. And so you have a lot of your familiar... It looks familiar to you except that now you have this chat window and this window that shows you the specifications that you can now interact with, and you can just start chatting. And at any time, you can say, "Hey, I want to do this as a specification." It's very easy. And one of the things that we did Kiro was you should be able to use Kiro within seconds of deciding you want to try it out. So you can just put a Gmail address in or a GitHub account and you're off to go.
John Furrier
>> Has Kiro gone into the HR business where it's like creating the coders?
Deepak Singh
>> Right now what it does is make the developer think .
John Furrier
>> Don't go there. It's the third rail. No, eventually coding will be... Code doesn't lie. Like you said, it's all on Git. What can we expect at re:Invent? Obviously agent is the hot area. We're seeing a mass acceleration. As the infrastructure gets faster, AI factories is the hottest thing. The build-outs are getting better. Trainium. You got the chips, you got this large-scale cluster that's powering a lot of action. What are we going to expect at re:Invent for agents?
Deepak Singh
>> I'll actually tell you a little bit about some of the things I'll be speaking about at my innovation session, and it's a lot about how the craft of software development is changing, how we are harnessing modern interfaces and the power of agents, and what are the things that we are doing to help developers bring the right context to an agent so that these really, really powerful agents that we are building that are enabled by the underlying foundation models are able to accomplish these tasks at velocities that none of us had seen before. And I like going back and talking about... Now I was not at AWS at the time. I still remember the first time I used EC2 and the way it blew my mind. We are at that stage now with software development where I have seen people take problems that they've avoided for years because it would just be too much work and over a weekend just get them done because they've learned how to work with these agents. So a lot of what you'll hear us talk about is what are the ways you can harness the power of these agents with Kiro and what Kiro allows you to do. And as agents get more powerful, as the LLMs underneath are more capable of thinking and reasoning, how should we, as developers, evolve to take full advantage, as opposed to using them once in a while for a problem, but make them the place where we live every day? And that's going to be a big part of our effort.
John Furrier
>> Yeah, it's so exciting. I can't wait to see the news and the talks. My final question for you is more of a personal one. I got my computer science degrees in the '80s. That was the systems revolution. Lived through the internet and the cloud. There's a lot of people older, over 40, that are coming back into coding and there's a lot of young folks coming in. Is there a new surge of... Because if you learn how to code and haven't coded in 20 years, you can just get back on the saddle. So the skills, and Matt Garment talks about this a lot with me about orchestrating and being the conductor or the auditor. I've done code reviews. I haven't done them in years, but I could get back in coding. So I'm hearing my friends come in saying, "I'm getting back in. I just built an app."
Deepak Singh
>> Yeah.
John Furrier
>> So this is-
Deepak Singh
>> The barrier to entry so much lower. My anecdote is, personal anecdote is, at home, I haven't written code for... Nobody should let me write code that gets shipped into production. But these days, anytime I want to show my middle schooler, explain a concept to him, I just write an app. We build an app together that we can then understand how planets move and how Newton's laws of motion work, and it's fun.
John Furrier
>> It might not have a billion users. Works great on a local host, but no, but this is the beautiful thing about the art, the craft. Craft is coming back.
Deepak Singh
>> It's fun. I've seen senior engineers who have written more code in the last six months than they had written in the three years previous, because these tools just allow them to express themselves in ways that they just couldn't before.
John Furrier
>> It's art. It's art. Code doesn't lie. Art's great. Deepak, thank you so much for being part of theCUBE special series here at the headquarters. Thanks for coming on.
Deepak Singh
>> Thanks for having me again.
John Furrier
>> I'm John Furrier with theCUBE. We are here in Seattle at the AWS headquarters at the re:Invent building, getting all the scoop for previewing re:Invent, but also finding out what's new. Thanks for watching.
John Furrier
Deepak Singh