00:31:49 · Teresa Carlson
A two way door, you could take it and you could come back. You still may have a cost and you still may have risk and you need to assess those, but okay. And so we looked at things from a one way and a two way door. And then the other component of this was really being able to put together a P and L, understanding not just the cost of risk, but what's the cost? Like, what is the actual cost? And how do we make sure that we can build it in the right way with the right cost structure? And then also we can make money. Like we're not looking at a losing game here. What's the ROI of this thing? So it was really, you had to be, I'll call a general manager. And what I mean by that is you had to know generally everything about your business. And that was very new to me. And while I'd written in my education, was a writer, I was more prolific in my writing in terms of words and I'm Southern, so sometimes, you know, I use fillers, which is my style. And so I really had to learn my writing style and my speaking style to be more crisp. And I used to coach my teams because we would have these OP-1s where you built out for the planning, what was your budget, what was your headcount asked, what were going to do for the year? And these documents, I would keep them up till three in the morning, rewrite it. We would reread it and rewrite it until I made sure that if I took that document in, it was something Andy, which was our leader, but in Jeff, if they read it, that he was like, I get it. And they could ask me the right questions and understand it. Another big lesson was if they ask you a yes or no question, answer yes or no. Like that was like another thing. People want to explain their answer before they give it. And I would coach my team once I learned. I was like, look, if you get a yes or no question, answer yes or no. And then if they want more details, they'll ask you. So there was these things you learned about our communication style. And these weren't harsh things. These were because we had a lot to do, we wanted to move through those things fast and people didn't want to waste time. The other couple of things I'll share that I really loved about what we did there was I built that business because they believed that I could do it, and they really backed me. And I think I'm a good example of how they built leaders.
00:34:18 · Teresa Carlson, Molly O'Shea
They would build a business around, and I did the same thing then. I would build a business. I built an aerospace and satellite business around a leader, General Clint Crozier that I hired, because I got him to agree to let me to build the business, and then I hired the right leader. And that business became a very solid business for AWS. But they believed in me, and I felt because they believed in me and gave me that opportunity to build it from scratch. And then they let me build out crazy things around the world because I delivered results. If I hadn't delivered results, it would have been another game. They also ask a lot of questions. I mean, it wasn't like, oh, you just get everything you want. You had to be very clear in what you're asking and what you're going to deliver. And as long as you did that, they were very open to you building out new business opportunities around the world. And I really appreciated that. And it taught me a lot about the art of the possible, what you can do and how you do it. But then once you're given that opportunity, you have to go deep, deliver results, have high ownership of what you've been given the opportunity to [00:35:33] S1: go do. Cloud was a huge transformational moment for technology. And now AI is in the same position. What are the parallels that you see between these two cycles?
00:35:44 · Teresa Carlson
[00:35:44] S0: The thing that has really, Molly, I have to say has really totally blown my mind is I think about we got started, my business, we build it from about 0 to 10,000,000,000 in the federal side or the government side globally in ten years. So that went really fast. I mean, was from nothing. And like I said, it wasn't, it was from us building up the market even. In AI, they've already way surpassed what they're doing. Like the numbers are off the charts. But I think for me, what surprised me the most about regulated industry and government, set commercial aside, how quickly that they're acquiring and adopting without the same restrictions that I told you about earlier with cloud, they still really have not defined AI. If you go look at what I said they define cloud, they've not really defined what AI is. And I think that is important. What is AI in the capacity in which it's going to be used? Because there's different models here still. You have your language, your model providers, and then you have your solution providers on top of that that are building AgenTic applied. And I think these things need to be understood by everybody, more education and training. So one is just they're adopting, they're buying without all these things in place. So while I'm excited about that, it also worries me because in my job as the CEO of the General Callis Institute, one of the early things that we came out with is we believe that federal government needed an AI framework. It was important to set some regulatory environment and put it in place for this. And I saw that with cloud. There were a lot of missteps in cloud. And so we had to back off that and really think about how are we training? Was there certification? And we were literally back then like going in and saying, look, you need to set a regulatory environment so companies also know how to meet this, not just large companies, but small companies. And I think for AI, I think it's fantastic. But I also worry that we're moving really fast without the frameworks in place.
00:37:59 · Teresa Carlson
And I believe Anthropic working with defense is a good example of if we would have had some framework that they were working off of, I think, you know, I don't know if the outcome would have been different, but it would have helped them as both the government and the commercial entity to have that conversation, and we just haven't gotten there. We still have a patchwork going on in these AI regulations, which makes it very hard on our startup community. But on just scale, I will say I used to tell my team at AWS, strap in because you won't see anything like this. And with AI, I'm like, this is a double or triple strap in. It's like nets. Prediction that I think that you'll see with AI is you're seeing a lot of spend, a lot of spend happening. And it needs to, you know, both with the model development and also just with customers' adoption, they're spending and they're using a lot of tokens. And I've shared with some of our founders that I believe you'll see what I call the optimization point, though. And we saw it with cloud, where there was a lot of spend. And then customers said, wait a minute, I want to optimize how I use. So this is where you saw the cloud providers reduce. They would have a model where we reduce pricing. Our motto was we wanted to decrease our price point. And then that would cause customers to spend less, but then they would adopt new services. So there was kind of a it was a real balance because we wanted that. We wanted the spend to basic compute and storage to go down so they could take advantage of other things like, you know, database services, streaming, AI, the other tools that we created, we wanted them to take on new services. So the spend over time actually kept going up, but you needed to reduce certain spend to see others go up because it's not an unlimited budget out there. And I do believe with cloud, you will see still a huge burst of spend, but then larger companies will say, well, wait a minute now, how do I optimize that spend? And what tokens, how do I really need to use it? And what workloads and workflows are the most critical to the success of my business with AI? Cloud evolved into a multi cloud environment, so you don't have dependency on one or the other and you can
00:40:29 · Molly O'Shea
[00:40:29] S1: have your data in multiple places in case there's an outage or something like Do you think something similar will occur in AI? Do you think people will stick to one model in the future or do you think we'll keep separate use cases for different things?
00:40:45 · Teresa Carlson
[00:40:45] S0: I think we'll keep separate use cases. You can't change human behavior. And human behavior, it comes to buying tech, is they're constantly looking. It's one of the it's a great thing for the industry, the tech industry overall. It's not always great for commercial industry. You see over and over companies that buy too much tech. And what I mean, they want to experiment a lot. And if they, then they kind of say, woah, I got to back to optimization. I got to look at all of the technology services I own. And am I really using them? This is just tech debt. This has been going on for years. And it takes companies a long time to get rid of their tech debt. And also if they've created an application on a platform, and then they're like, well, now I've got to move it. And that becomes very challenging, especially for large enterprises or governments. But in AI, yeah, I predict that we are so early in what is being created and how quickly it's being created. I think our, first of all, we're trouble processing just what's happened in the last few years, but it's even going to go faster. So you're going to see more models created, you're going to see more solutions created, more adoption because as humans, we're going to learn as leaders, we're going to really determine what's the best model we use with our employees to teach them how to take on AI. We're very early with that. Most companies are still trying to determine, can I use it at all? How do I teach my employees? Am I okay with them using AI? I mean, even my team still is like, are you okay? I'm like, go use it. You have to make sure that you are doing background checking, that you're checking your data. I mean, it still hallucinates a lot and it tells you even, hey, check my work. So we're still at the point we need to do a lot of checking, but you've got to get your employees really okay using it. And now as new employees come online, they've got to have a lot of agency to both learn AI and understand how they manage with AI. So there's kind of two different models. But I think you're going to see workloads go in all different directions with all different companies' uses. And it's going to be who's the leader at the time.