Amid the breathless coverage and relentless AI hype of recent years, one of the world’s biggest tech companies—Amazon—has been notably absent.
Matt Garman, the CEO of Amazon Web Services, is looking to change that. At the recent AWS re:Invent conference, Garman announced a bunch of frontier AI models, as well as a tool designed to let AWS customers build models of their own. That tool, Nova Forge, allows companies to engage in what’s known as custom pretraining—adding their data in the process of building a base model—which should allow for vastly more customized models that suit a given company’s needs. Sure, it doesn’t quite have the sexiness of a Sora 2 announcement, but that’s not Garman’s goal: He’s less interested in mass consumer use of AI and more interested in enterprise solutions that’ll integrate AI into all of AWS’s offerings—and have a material impact on a corporate P&L.
For this week’s episode of The Big Interview, I caught up with Garman after AWS re:Invent to talk about what the company announced, whether he feels behind in the AI race, how he thinks about managing huge teams (and managing internal dissent), and why he’s not convinced that AI is (or should be) the great job thief of our era. Here’s our conversation.
This interview has been edited for length and clarity.
KATIE DRUMMOND: Matt Garman, welcome to the Big Interview.
MATT GARMAN: Thank you. Thanks for having me.
We always start these conversations with some very quick questions, like a warmup. Are you ready?
Go ahead. Fire away.
If AWS had a mascot, what would it be?
We have a big S3 bucket sometimes that goes around, so we’ll call it that.
Sorry, what is an S3 bucket?
An S3 bucket is like a thing that you store your S3 objects in, but we actually have a large foam big bucket that walks around and actually looks like a paint bucket.
So you do have a mascot.
Well, S3 has a bucket, it has a mascot. It’s probably the closest we have, and I like it.
What’s the most expensive mistake you’ve ever made?
Personally or professionally? That’s a good question. Personally, the most expensive mistake I ever made was playing basketball too long and I tore my Achilles. So that cost me about nine months of being able to walk. I probably should have known that into my thirties I was well past basketball-playing age. I lost a little bit of time there.
That sounds personally expensive.
Yes.
If you could rename the cloud today, what would you call it?
What is it called today?
The cloud.
I actually think the cloud is a pretty good name. So I don’t know if I would rename it. I would rename what we called Amazon AWS, or Amazon Web Services. Now, no one knows what web services are. So that, I might rename a little bit, but I actually like the name of the cloud, so I’m not sure I would rename it.
Maybe Amazon Cloud Services.
Yeah, maybe.
What’s the part of your job you would love to outsource to an AI agent?
I try to outsource a lot of the parts of my jobs to AI agents, if I can. But I find that I still haven’t figured out how to outsource answering my day-to-day emails yet. That still, I find, takes up a lot of my time. If I could figure that out, I think that would be great.
But it sounds like you’ve tried, and I know we’re supposed to be doing quick questions, but I was going to ask about this later. Tell me a bit about how you’ve tried to incorporate artificial intelligence into your workflow, into your personal and professional life.
There’s a number of ways that I’ve done it. In particular for me, though, a lot of the benefits that I get in my job are taking a lot of information inputs and then kind of sharing those out to either the same or other people and connecting a lot of those dots.
For my particular role, I haven’t yet found a huge shortcut for being able to do that, particularly with regards to the medium in which we communicate, like email or other places like that.
Because I find that all of the shortcuts lose some of that nuance. There’s some summaries and things that work. There’s definitely some tools that allow me to summarize content more quickly or learn new content more quickly, which I find to be super useful. But I haven’t found a huge time win for my role in particular, where there’s not as much repetitive work or other things like that.
It largely is knowledge that I’m trying to get from a bunch of sources and then consolidate to send out to others.
That’s actually very interesting to me, and reassuring in a way, because I sometimes feel like I should have found a bunch of shortcuts by now.
So, from one former AWS intern—that’s you—to a future one, what is your best piece of advice?
I find that people always overestimate how much technology has already been invented and think that there’s nothing left to do. What I find is that we continue to be at the early stages of evolution. As long as you’re curious and looking and willing to try new technologies, new areas, we’re always at that early stage of what can be invented.
Sometimes I run into interns who are like, “When you started AWS it was small, but now it’s a big company and there’s not the same opportunity.” And I’d say that that’s just not true. I think there’s just as much, if not more, opportunity than there ever has been.
Which leads me to my next question: How do people know when you’re unimpressed?
When I’m unimpressed, I would tell them directly.
Fair. So, direct feedback.
Largely, though, it’s not about impressing me. I like when people are thoughtful, when they’ve come up with the right sets of decisions. It’s not like people are coming up with something that’s super novel every day. I like it when people have done the work so that we have all the information, whether it’s customer input or data input or sales input or whatever, so that as a group we can make thoughtful decisions.
I’m impressed when people have done that work ahead of time so that when we do get together, we can make good decisions, and thoughtful decisions. I’ll often tell people it’s not necessarily the recommendation that I’m going to be impressed by, but I want us to have all the information so that as a team, we can move forward and make great decisions.
What is a hobby you wish you had more time for? I’m assuming it’s not basketball, based on what we just learned about you.
It’s not; I’ve switched to golf. I caught the bug a couple years ago and quite love playing golf. I don’t get to play as much, but I enjoy it.
Let me set the stage a little bit. We’re here to talk, in particular, about a bunch of announcements that you recently made around AWS and AI and agentic AI that WIRED covered. I’m biased, but I thought we did a great job with that coverage. But I also want to learn a little about you in a professional context. So tell me, and tell all of us, about your career journey thus far. Now you’re the CEO of AWS, but you had a very long career at Amazon before that. How did you end up where you are now?
I’ll start at the beginning. I worked for a couple of startups early on in my career. None of them did particularly well, but I learned a ton from them, which was great. After my second startup, my wife and I both quit our jobs and went to business school, which was a fantastic opportunity.
When I was there, one of the things during my internship that I wanted to try to explore was what entrepreneurship looked like inside of a company. My goal was always to go back and do a startup again. So as part of that, I looked at a bunch of different companies and I ran across Amazon and actually talked to [president and CEO] Andy Jassy, and they talked about building this technology services capability inside of Amazon. I thought that was exactly what I wanted to see. I wanted to see what it would look like for a successful technology company to try to build something new inside of it, because I just wanted to see what that motion looked like and how I could learn from experienced entrepreneurs and what was different than at a startup.
Got it.
So I did my internship for what turned into AWS in 2005 before we launched. I was fascinated by it. I thought it was an awesome opportunity, and I said, “Great, I wanna come back and work here for a couple years,” and then I would go back and do a startup.
So then I started full-time in 2006, effectively as the product manager for all of AWS. It was largely defining all of the services as we launched them. So I started a couple weeks after S3 launched, and before the rest of our services launched, and helped launch them and name them and price them and do a bunch of things.
Then I kept getting more and more responsibility. I focused on EC2, which was our compute service. I started taking on engineering teams, actually launched our block storage service, kind of wrote the PR/FAQ for that and hired the first engineering team and launched that.
Then I grew to lead most of our core compute and networking and storage product areas. So all of the product and engineering teams for that. It was fun. I got to learn a lot along the way. I’m not necessarily, or I wasn’t originally, a deep technology person, but I got to learn about hypervisors and kernel engineering and a bunch of these really low-level core technology pieces, which was cool. It was super interesting. It’s just such a fascinating space, and as AWS grew really rapidly, we grew along with it, and we grew the team pretty significantly.
We were fortunate enough to work with some of the best technology people in the world as we built the service, and a bunch of services, as the business group. I can’t remember the exact time, but it was after about 12 or 13 years, so it would’ve been like 2019, something like that, Andy Jassy called me in his office one day and asked if I would lead sales and marketing. I literally didn’t know anything about sales and marketing. In fact, I was kind of looking around to see if he was talking to someone else. But it was a great opportunity to learn that space, and it was a unique opportunity, right?
Yes.
I was basically handed one of the world’s two or three biggest enterprise sales and marketing organizations, having never done any of those jobs before. I think Amazon’s a bit unique in that we trust people. If you’re smart, you know how to operate, you know the business, you don’t necessarily have to know the exact thing that you’re going into. The team was gracious and helped me learn that. That was a great opportunity to get to learn how to run a field organization at scale and get to spend a ton more time with customers, to really understand the nuances of what a startup customer was looking for versus enterprise, versus the governments, which was awesome.
Then I took over as CEO two years ago, or a year and a half ago. So I’ve spent almost 20 years here at Amazon, all in AWS.
How big is the organization by employee size?
I don’t know the exact numbers, but it’s in the hundreds of thousands. We have large data centers. We run a very large operational organization. We have data centers all around the world. So it’s a pretty big team.
I love management, and it was clear fairly early in my career that that is where I wanted to go. Was that always clear for you, the idea that you wanted to be taking on more and more of the enterprise? That you were doing what you were meant to be doing?
Yes, I like it and I think that I’m reasonably good at it, I guess.
We’ll ask some of your employees.
I took on more and more as it went. You can ask others. I guess that’s not for me to say, but I liked it. You know, it was a ton of learning. What I really love is when I get to take on roles where I get to learn more, and get to stretch myself.
Frankly, I love building and love having an impact on what the business is doing and what our customers are doing. It’s hard because you don’t get some of the joy of actually physically getting to build the thing or deliver the thing yourself.
In my world, that’s like not getting to write the story.
That’s right. But you can write a lot more stories in that case. And you learn to do that through others, which is an interesting and useful skill. You learn how to communicate to teams, first through direct management, then through layers of management.
So you gotta think about how you build mechanisms. This is one of the things I enjoyed learning, which is how do you think about building mechanisms that allow you to help those individual contributors make some of the right kinds of decisions, and the strategy that you’re trying to drive for the team or the business or the company? I’ve quite enjoyed learning how to leverage some of those mechanisms at different scale. That’s been fun to do, too.
You just talked about going from managing people to managing managers, and I swear I’m asking for a friend, but do you have any particular mechanisms that you have picked up in those 20 years that stand out to you as particularly effective strategies? Because I will say there is something very specifically different about managing someone who then manages people who then manage teams of people.
Yeah.
It has the potential to be a very unproductive game of telephone. What mechanisms do you employ to make it much more effective than that?
I mean, look, everybody has their own way of doing this. So you a little bit have to find what works, and I do think that as your team in your organization gets larger, you have to change some of those things. I think that’s one of the common pitfalls that I see people fall into is that they will assume that the thing that worked great when they were a line manager, managing a team of six to 10 people, will work the same as when they’re managing a team of a hundred people.
Those same things won’t work. There’s another shift that’s like when you don’t know the name of everyone in your team. Or your organization, which I can’t unfortunately know today. Usually that breaks somewhere around a hundred to 200 people, somewhere in there.
Right.
You’ll run into people who are in your team that you don’t know or don’t know their names. You just have to think about all of those things differently. You want to give the broadest set of people in your team as possible mental models on how you would make decisions in their place as opposed to what the decision actually is.
Because if you send down edicts all day—like, we’re going to do this, or you wanna do this, or I’m going to make a decision on that—it’s not as empowering to your team. And there’s a chance that whatever you say gets miscommunicated. But if you can give people mental models of, for example, “If I was going to approach this situation, this is how I would think about it, or these are the ways in which I would make trade-offs,” then you can empower tens of thousands of people to make decisions.
Then you can focus on making sure that you hire smart people and don’t punish them when they make bad decisions, but instead course-correct. That’s how I’ve kind of learned at scale, where your real leverage points are ensuring that you have those right mechanisms at place.
I appreciate the free management advice. I’m sure many of our listeners do, too, so thank you.
I want to now ask you about AWS in a broad context. I would describe WIRED’s audience as sort of curious generalists. Obviously, they’re listening to this, or reading this, because they’re interested in technology and where it’s taking the world. They probably have some sense of what AWS is. But they probably don’t know how big AWS is and how vital it is in terms of just infrastructure. Can you explain it to us?
Not like we’re five, but like we’re 21, we just graduated with a humanities degree, trying to understand this thing that you are in charge of.
At a high level, our idea behind AWS hasn’t changed in the last 20 years, which is that there are a bunch of pieces of technology that were really non-differentiating that companies had to do on their own for a long time. It used to be that they had to build a data center. They had to go find servers, they had to take care of the servers. When a disk drive broke, they had to go fix it. They had to set up their networks, et cetera. There’s a whole bunch of work that they had to do before they could ever write a cool application that would let them stream a movie or connect individuals with people who had rooms they wanted to rent out.
When you think about all of that work, our goal was: What if we could do that work for companies so that they didn’t have to do that? That was the thesis when we started. So that someone could come and say, “Great, give me servers, give me storage, give me databases,” and we’d provision it for them. Then, through an internet connection, they could have access to that. It turns out that was a very powerful idea. I think we did a good job executing on some of the abstractions that made it really powerful.
Netflix and Airbnb and Pinterest are all some of these early customers that we had that built their business from the beginning on AWS and the cloud. They don’t own data centers, they largely just run inside of AWS. Initially, when we first launched the business, we thought this was going to be incredibly compelling for startups and some technology companies, which it was, but as we grew, we found out that enterprises and really large organizations were equally compelled by this value proposition, and we had to build a lot more capabilities for them. Whether it was like encryption capabilities or audit logs or abilities to hit particular compliance things or whatever it is.
Got it.
But now we have customers like Pfizer and JP Morgan, and the United States government and the intelligence agencies. That was a big win for us when we kind of convinced the US government that we could build a top-secret region and that they could run intelligence workloads inside of AWS.
I would’ve loved to have been a fly on the wall in those meetings. What does it take to convince the United States government that they should run on the back of AWS?
We walked through some architectural questions that they had, and … you know, there was a whole [request for proposal] process and there’s a lot of work that we did, but it was also just some whiteboarding where we kind of walked through how it would work, and at the end of the day, the cloud sounds like it’s a magical technology, but it’s data centers and it’s networks and it’s servers, and other things that we run at very high reliability, at very high security.
We found some forward-leaning technology folks that wanted to figure out how they could get the benefits to the government. It turns out that if you can find the right person in an organization, even somewhere as large and bureaucratic as the US government, you’ll find people who want to lean forward, go faster—they want to deliver value for citizens. So, finding that right person and then being able to collaborate with them, their eyes light up just like they do for a startup company, right?
These people are oftentimes very mission-driven, and if they can see an opportunity to deliver more value for the mission that they’re on they’re super excited about that.
So today, now it’s across almost every country and every industry that you think about throughout your daily life. There’s still lots more to go, but I would say there’s a bunch of companies that now use AWS as their core infrastructure. NASDAQ trading markets run on AWS; financial services companies run on AWS; hospitals run on AWS; media and entertainment—live broadcasting or streaming—we power a lot of that technology.
Just when you thought you were done learning the job, along comes artificial intelligence. Which brings me to my next question. Obviously, we are in this new era for you and your organization. You held this re:Invent conference in December, and you streamed it on Fortnite, I might add …
Yep.
You announced some pretty significant changes to AWS, to your mission, to your priorities, and to what you would be offering to your consumers. I’m hoping you can talk us through that transformation.
Technology is always iterating and going through transformations. So for us staying at the forefront of every innovation is incredibly important. I think there’s been almost no technology leap since maybe the cloud, and the internet before that, [as big as AI]. So we’ve been investing in AI, in AWS and Amazon, for the last decade-plus, two decades maybe. But definitely with the leap forward from generative AI over the last three years, we’ve just seen a massive change in what’s possible for customers.
So we’ve had this vision that it’s not just going to be, AI is over on one side and then the rest of your business is going to be over on the other side. Basically AI is going to be built into what everyone does. In order for that to happen, number one is you have to have all of your data in the cloud world. And we’ve built this whole platform of tools that then allow you, once you have that data in the cloud, to deliver differentiated value to your customers.
Some of the things that we launched—that I’m quite excited about—at re:Invent are really around AI agents and the difference between the first generation of AI tools [and the next]. The first generation were really around summarization and content creation, right? We were all quite excited and got a lot of value out of those. But there’s only so far that goes. I think the next stage is these agents. The real value is they can take access to your data—they can still do some summarization and content creation, but they can actually accomplish tasks. They’re able to reason.
When you have these agents that can reason and accomplish tasks on your behalf, all of a sudden you can kind of force-multiply what you’re able to do. So we launched a couple of things. One was called Nova Forge, where we allow customers to actually take their data and integrate it in at the early stages of training one of these frontier models. So that gives enterprises one of these AI models that deeply understands their data and their domain. Then we launched a number of these frontier agents that allow customers to really deliver big bodies of work, whether it’s in coding or operations or security.
We’ve spent the time to really build this platform where it could actually deliver value. I think broadly people have understood what that long-term strategy was. They really get it. As they see projects really delivering, where there’s real value, they see that AWS is the platform where they want to go do that. That’s what customers were telling us over and over again [at re:Invent].
I feel like 2025—which we are, thank God, almost done with—was a confusing year in the narrative for AI. I say that because I feel like in January I went to some conferences, talked to some people, and the sentiment was “This is the year of the agent. Agentic AI is here. It’s about to change everything.” That was almost a year ago.
Yeah.
Am I using agentic AI right now? Absolutely not. Has that come to fruition in the way people were talking about in January? No. At the same time, you’ve seen reports from MIT, for example, finding that 95 percent of gen AI pilots in companies are failing to yield the productivity gains that I think those corporate leaders thought they would see.
How do you make sense of all of those narratives coming together? Were companies just moving too quickly?
If you jump forward and don’t build that strong foundation of I have my data, I know the workflows, I really know how some of these things are going to tie together, then you’re not going to get any value out of it.
You’re going to have just a chatbot, which is kind of cool and looks neat. Then everybody has a chatbot and then what? So it is those differentiated workflows, you know, and they’re less sexy and interesting for people to look at, but a workflow that can help you automate insurance claim processing is super valuable, right? You can actually get people’s claims processed faster, make sure that you cut your costs, have a better customer experience, make sure you have better accuracy—you can deliver all these things.
Some of the technologies weren’t available in January. This technology is moving so fast that the capabilities are much better today.
I will tell you, at re:Invent I sat in a room of executives for a broad set of companies, and I asked for a show of hands of who is either now starting to see positive ROI on their AI investments or see a clear path to meaningful positive ROI in the next six months. I think 90 percent of the hands went up.
Hmm.
That’s not the answer I would’ve gotten a year ago, but it’s because we’ve done a bunch of this work, and it’s because we’ve done a bunch of this work together with customers to understand exactly what they want, how do we solve their problems, and how do we deliver solutions that deliver them real value and not just, you know, clickbait headlines that sound good.
On that note, I will say candidly, Amazon has not been a key part of a lot of these AI narratives, right? There have been other companies that have been out there making announcements. I can tell you, leading WIRED, it’s this constant stream of news, “We’re doing this, we’re doing that.” Does that worry you? Do you worry about Amazon not being in that narrative, or do you feel like you took your time?
I think both of those things are true. I do worry about it, because I don’t want customers to think we’re not innovating or driving the latest technologies they need. I want to make sure that it’s not just headline-grabbing stuff and it’s actually great.
Jeff Bezos used to have a saying that you have to be willing to be misunderstood for long periods of time. For us, I think that’s what some of the last two years entailed.
I think a lot of that narrative has changed now. If you talk to lots of analysts, if you talk to folks in the press, if I talk to customers they’re saying, “Look, actually AWS has by far the strongest agentic platform to go build on. They have the broadest set of models that I can build on. They have the broadest set of security controls and compliance controls that actually, if I go put these agents in production, I can actually audit, know what they’re doing, control what they’re doing.” That is what we’re seeing.
That’s not to take anything away from [other AI tools]. ChatGPT is an incredible consumer application, but it’s just a different thing. That’s not our business. Our business is to make sure that banks and health care companies and media and entertainment companies and energy companies can drive their businesses and deliver more outcomes for their customers or cut costs or whatever they want to do.
So, I’m quite pleased with where we are now, and I think we’ve already seen that narrative largely shift.
I wanted to ask you a little bit more about Nova Forge. What was particularly interesting to me here is this idea of custom pretraining, as opposed to the idea of fine tuning, as a way that a company can take a model and really make it their own. Can you explain that distinction to everybody before I dig in a little more?
It’s not new that people have thought, “OK, there’s these out-of-the-box models that I wanna customize,” and the best mechanism that we’ve had to date to customize are these open-weights models. Meta was great and released that with their first Llama model, but now we’ve seen a variety of these, whether it’s Mistral or DeepSeek or Qwen or whatever.
There’s a number of these, but they’re still black boxes, right? You basically get a fully pretrained model, and then they open the weights and you can tune the weights to focus in particular areas. Or there’s new techniques like reinforcement learning where you can start to send more information to these models that train them after the fact.
What we find is that if you put too much new data into these models in the later stages, they forget the early stuff and so they forget what made them great at reasoning. Or they actually forget some of that data because they get what’s called overtrained on some of the data you’re giving it.
Right.
Then they lose what was valuable in the first place. So there’s only so far that that can go. What we also find is that those techniques are very ineffective if the model wasn’t already trained on your domain. So if you try to teach one of these open-weight models about protein folding and they know nothing about protein folding, it doesn’t work, because it doesn’t know how to inherently reason about that thing.
What we found is that if you can train them earlier, and I use this analogy in my re:Invent talk about the human brain, it’s like learning languages. You’re able to learn new languages early when you’re younger. Much easier. Now, if I try to learn a new language, it’s much harder to do. So models are somewhat similar to that.
What we’ve done, which is a unique thing, we refer to it as open training, but really the idea is that you have this corpus of data that is from your domain and your particular company and the ways that you do things. If you’re able to insert it into the pretraining stages and then mix in a bunch of the data that was used to originally train that model and then finish pretraining the model, when it’s done the model actually inherently knows all of your stuff. It knows about your data, it knows about your company, it knows about your domain, and anything that you do with fine-tuning or post-training after that is actually much more effective.
It’s just never been possible before. Because open-weight models never would expose their data, and there was no kind of mechanism for them to be able to go do this. So we did this with our Nova models and our Nova 2 models.
Take a financial services company. You can take all your information, inject your data, mix it with an Amazon-curated dataset. This is the data that we have proprietary use. We’ll give you tools to easily mix that together. And then you finish pretraining the model, and you effectively have your own custom frontier model that understands your business that you were able to train for a couple hundred thousand dollars or whatever the cost is to finish that training, versus billions of dollars of doing all the research to actually go and build a frontier model.
I’m curious about risk in the context of Nova Forge in a few different ways. You already deal with companies that have confidential proprietary information, right? You just mentioned financial services inputting all of that data into this model and into this training. You also offered a really compelling example off the back of re:Invent from Reddit, which is using Nova Forge to develop a model that can be used for content moderation, which essentially means training a model that breaks a lot of conventional rules around how these models are typically designed, right? They would be designed to avoid offensive or violent content entirely. Reddit, on the other hand, needs a model that can lean into that kind of content in order to do the moderation. So those are two very distinct examples that I just offered, but I’m curious what kind of risks exist when we start down the road of this kind of custom pretraining, and are there distinct risks that might be different from what we’ve seen historically?
I don’t know if there’s any particular, different risks. I think from a data protection point of view, we have all sorts of protections around this. That’s one of the things we pride ourselves on, and have over the last 20 years, is really protecting customer data and making sure that it’s isolated and protected.
When people are building their own model, this is true no matter what, if they’re fine-tuning it, if they’re doing any of that work, companies have to think about what is the output, and they have to own the output of their own models.
Mmm-hmm.
Even if they’re using an off-the-shelf model, by the way, you have to own the outputs of that.
I think that’s one of the key pieces: You can’t just give up responsibility for whatever your technology is doing. You can’t give up responsibility because a database makes a particular join and you’re like, “I don’t know, it’s just how the database did it.” AI is no different from that.
We deliver powerful tools to customers, but they have to own those outputs and think about them. We still have safety classifiers, by the way, for things. You can’t pretrain something and then create it to build a bomb or things like that. We still have a lot of those safety controls, and there’s no circumventing those.
But with regards to things like content moderation, that’s someone else’s choice, and they can make a choice and they own what that looks like. They have to be sure that they make great choices for the content that’s on their website, that’s appropriate for their site.
So, we’re talking about this premise whereby AI agents will be much more integrated into enterprise settings, and I’m curious how you think about that in the context of the workforce. I’m looking back at 2025 and remembering comments that other AI executives have made about job disruption, cuts to the workforce, et cetera.
You actually made an interesting comment where you said that replacing junior employees with AI is, quote, “one of the dumbest ideas” you’ve ever heard, which made me laugh.
Ha!
Could you talk more about that and more about how you see artificial intelligence and agentic AI changing the workplace in the years to come. Because I think you have a point of view on this that some people might find reassuring, and that I think is different from what we hear from a lot of other leaders.
That point, by the way, was specifically about software developers, but I think it applies to lots. There was this thought that you’ll just replace all of your junior engineers and all of your junior employees and you’ll just have the most senior, most experienced employees and then agents.
Number one, my experience is that many of the most junior folks are actually the most experienced with the AI tools. So they’re actually most able to get the most out of them. Number two, they’re usually the least expensive because they’re right out of college and they generally make less. So if you’re thinking about cost optimization, they’re not the only people you would want to optimize around.
Three, at some point that whole thing explodes on itself. If you have no talent pipeline that you’re building and no junior people that you’re mentoring and bringing up through the company, we often find that that’s where we get some of the best ideas.
We get the new fresh blood into the company from hires out of college. There’s a lot of excitement. There’s a lot of new thoughts. You’ve gotta think longer term about the health of a company, and just saying “OK great, we’re never going to hire junior people anymore,” that’s just a nonstarter for anyone who’s trying to build a long-term company.
What does this mean, though, for the workforce broadly? What does it mean for Amazon’s workforce? What does this look like as AI agents infiltrate—which is not a generous word, but it’s the word that comes to mind—the way we work and live.
One of the things that I tell our own employees is “Your job is going to change.” There’s no two ways about it. If there’s only one thing I can promise you, it is that the way that you did your job four years ago is not how you’re going to do your job next year. You’re going to be able to have a bigger impact. You’re going to be able to do more things, you’re going to be able to have a broader scope of responsibilities, and it’s just not going to be the same things that made you successful five years ago. You’re going to have to learn new skills, you’re going to have to learn new ways of working. We may have to organize our teams differently. We may have to go after problems differently. So people are going to need to be flexible. There is for sure going to be disruption in how work is done, because jobs are going to change and industries are going to change.
If they don’t, they’ll most likely get left behind by people who move faster and do change. There is going to be some disruption in there for sure. Like there is no question in my mind.
When you say disruption, do you mean in the context of job loss, in a big picture, economic way?
That is uncertain to me. I’m very confident in the medium to longer term that AI will definitely create more jobs than it removes at first. I think anytime you find opportunities to create new economic prosperity or build new experiences or things like that, there are more jobs that get created. But they will be different. There are jobs that will be eliminated as part of it, or reduced, almost for sure.
There are some things where you just no longer need quite as many people to do a particular job. So our job is to also provide training and upskilling so that we can retrain, so that there are other roles that some of those folks can do. Not all people want to do that. There may be some churn in the short term, as people either are hesitant to learn new skills or don’t wanna learn new skills or other things like that.
But this is true of almost every single technology change: It was true of industrial automation in the 1930s. It was true when personal computers came around, and when the internet came. It will be true for AI too. There’ll be some jobs that there won’t be as many of, that is true. And I think there’ll be new jobs.
I’m curious how you’re thinking about all of this from an environmental perspective. I think one of the notable pieces of agentic AI is having them run continuously for hours or days. I would imagine there is a sustained energy demand in that context. We’re already talking about a very energy-intensive technology. Amazon right now is the single biggest purchaser of new renewable energy contracts in the world, at least for the last five years.
That’s right.
How do you put that commitment together with this drastically increased need for energy?
Look, part of what that is, is us investing. This is not just going out and buying existing things. That is investing in new projects and bringing new renewable energy projects online. That’s a huge commitment for us. It’s something we do every single year and we’ll continue to do. Because we think that’s important from an environmental impact point-of-view, and frankly, from an energy availability point-of-view. I do think we’re going to have to keep looking at other energy sources. I think that nuclear power is one of those that’s going to be very important for us to look at in the medium to longer term.
We have to make sure that we have enough carbon-zero energy out there, but we need to look at all of those sources of energy, and it doesn’t mean that no one’s going to use natural gas in the inter-medium term. I’m sure some of that is true, too. Our goal is, how do we continuously, period over period, year over year, reduce the carbon intensity of the energy that we consume? We’re still committed to that as a company, and we’ve spent an enormous amount of time on that.
How do you respond to criticism of this venture, which is an enormous undertaking, internally? I’m curious about the management piece there. A few weeks ago you had over a thousand Amazon employees—granted, it’s a company with a seven-figure employee base, if I’m correct—but they described the company’s quote, “all-costs-justified” warp-speed approach to AI development. They say that it will cause, quote, “staggering damage to democracy, to our jobs, and to the Earth.” That’s quite a statement from Amazon’s own employees. When you hear that read back to you, how do you address it?
Number one, we encourage our employees to have their own thoughts as to how they’re thinking about things.
Well, that’s good because a lot of tech companies these days are not happy to see that happen.
I would also say that when you have an organization of any size, you have viewpoints from a lot of different places. I would say that that is not the majority opinion from our employees or even close. I think most of our employees are excited about the technology we’re building. Excited about the value that we’re giving customers, and excited about the potential and the climate pledge that we have and the path that we’re on. So, I think that’s OK. As long as those concerns are respectful, we’re willing to listen to them. But I think they’re far from the majority.
I mentioned at the top of the conversation that in January 2025 everyone promised me this was the year of agentic AI. They were wrong. And so as we look out to 2026, I’m curious for your prediction: In the context of AI, what is next year all about for us? Then in 12 months, I’m going to call you and we’ll see.
Just be super, super clear, I’m awful at predicting the future.
I am too, but I have to ask.
You could call me, I just probably won’t be right. But I will say I think one of the big things that we are going to be incredibly focused on for our customers is delivering real business returns for them. Whether that’s through agents, whether that’s through customized models, whether that’s through scalable infrastructure, whether that’s eliminating tech debt through using [Amazon’s] Transform to get them off of mainframes or get them off of legacy databases. That, I think, is going to be a big focus. It’s not going to be about going and experimenting. It’s not going and trying new technology. It is really delivering value to the end customers and to the business.
So it’s time for the P&L to look a little bit different next year.
Yep. I think that’s where customers have relied on AWS to help them improve for the last 20 years. And it’s what we’re particularly great at.
How to Listen
You can always listen to this week’s podcast through the audio player on this page, but if you want to subscribe for free to get every episode, here’s how:
If you’re on an iPhone or iPad, open the app called Podcasts, or just tap this link. You can also download an app like Overcast or Pocket Casts and search for “Uncanny Valley.” We’re on Spotify too.
The post AWS CEO Matt Garman Doesn’t Think AI Should Replace Junior Devs appeared first on Wired.
