Software Designing & Software Development Is An Art In Itself | Mark Schwartz | AWS | AIM TV

I am not yet convinced that AI can perceive the beauty of something. Can it write beautiful code? Maybe.

I was a government CIO for a while, wanted to transform the organization and be able to be really responsive to change. If I had had the AI tools available to me at that time, it would have helped me.

Oh, thank you so much. It's a pleasure to be here. How are you doing?

Actually, software came first. I studied computer science at university, and I was always doing it because even when I was doing film and theater, I still needed to make some money on the side. So I was leading software development projects and that kind of thing. If you've done a lot of software engineering, you might be able to see it as a kind of art. It's an aesthetic thing: you write a perfect program and it's beautiful, and you know it even if nobody else knows it.

Well, I am not yet convinced that AI can perceive the beauty of something in the same way that a human being can. Can it write beautiful code? Maybe. AI does amazing things with photos these days and video photographic manipulation, and what it's producing could in some sense be art, or at least in many ways is like art. Whether it actually perceives it, I have a little trouble with that right now.

Oh, I like to see high quality work of any sort. I guess that sort of holds together a lot of what I believe. I'm interested in high quality literature, good literature, but I'm also very open. Sometimes people say what kind of food do you like, and I say any kind of food as long as it's really well done. As long as it's good food, as long as it's really good, I'm open. So I would say in a sense, yes. There's another parallel that maybe isn't so obvious between theater and film and computer science and software engineering on the other hand, which has to do with management. If you're directing actors, how do you do it well? You don't do it by saying move your left arm a little bit, then move your right arm. You can't micromanage it. Yet as a director, you want your vision realized on the stage. So somehow you have to convey it to the actors, and it's never obvious how to do that. You have to invent and innovate ways to communicate with the actors. Sometimes it was having them do exercises and games, sometimes it was me leaping on the stage and playing opposite them and doing something. When you're dealing with software engineers, the really good ones are prima donnas. You can't tell them exactly what to do, and they know better than you, and in fact they do. So how do you get the result that you want? Sometimes you have to be a clever, creative manager to figure out how to do it, and I think I learned a lot about how to do that from managing actors, directing actors.

I always see it as working backwards from the result that you're trying to get. How do you get that result? You have a lot of contributors to the result. Some of them are people. In theater, lighting for example and set design, you have to somehow coordinate them and bring them together. Managing each of them is slightly different. You manage people one way, you manage your lighting designer a different way. So if you're going to be working with generative AI or other types of AI, you're going to be working with people, you're going to be working with different technologies, you're going to be working with peers and people senior in the organization. You have to orchestrate it all somehow to get the result that you want. To me, that's the interesting and hard task of being a leader.

I have maybe a few different ways to answer that. The first thing is when you're trying to do something difficult, and entrepreneurial ventures are always difficult, you need a source of energy. Something has to be pushing hard and injecting energy into what's going on. That's the role of the entrepreneur, that's the role of the leader. I was a government CIO for a while, in homeland security, so it's quite bureaucratic, let's say. Getting things done was really hard. It used to take us five to ten years to do an IT project, and I wanted to transform the organization and be able to be really responsive to change. The challenge for most leaders in the government is they might have in their head what they're looking for, they might have a lot of people work for them who want the same thing, but it doesn't go anywhere because there's no push, no drive. There isn't somebody who keeps asking hard questions and saying why can't we do this faster, and going and negotiating with other people. So I see the entrepreneur as a leader who injects energy, and that's the most important thing to me. But I think there's another aspect to it: I haven't found leaders to be successful if they're too prescriptive. If you have a lot of really good, smart people working for you, you want to take advantage of them, involve them in making good decisions and coming up with innovative ideas. But you don't just want good ideas, that doesn't help. They have to be good ideas that are actually going to move the business forward. So the role of the entrepreneur that I see is to empower and enable the people working for them, and at the same time mold what they're getting from those people so that it actually makes sense in the context of what they're trying to do. So maybe I've given you two answers to your question, both of them.

Yeah, and you make it sound like there was actually some logic behind the whole thing. It was all just take it as it comes, new opportunity, it was interesting. So I was in the federal government. There's nothing funnier than the US government once you're in it. Imagine this: one day I was watching television, I was watching the news, and President Obama at the time came on. He was giving a press conference and he announced this big new initiative, a big immigration initiative called DACA, Deferred Action for Childhood Arrivals, that turned out to be his big thing. It was a big change, and he announced that it was going to be rolled out in 60 days. Now this was a big shock to me because actually I was the CIO for immigration, and it was my team that was going to have to roll it out in 60 days. This was the first time I'd heard of it, and he hadn't asked us if it was possible, of course. I knew a few things that the president didn't know. If he had asked me, I would have told him, but he didn't know that in order to launch this we were going to have to make changes to about 25 big legacy IT systems, and our average time for making even a small change to an IT system was 18 months. So the math didn't work, obviously we could not do it. And obviously we did do it because he was the president and he said so, but it was bad. Nobody should do things the way we had to do it. In those 60 days, we got it done, but I had people on my staff who didn't sleep, and maybe we didn't completely pay attention to our security rules. We just had to do whatever we could to get it done in 60 days. But the lesson that I took away from it is immigration is going to change a lot, and we're going to have to make a lot of changes and respond to political needs. So saying that it's going to take us 18 months to do something is just not acceptable. We have to find a way to do it much more quickly. So that is what we focused on. I became all about speed. I said how can we, instead of taking 18 months to make a change, do 100 changes every day? In the end, that's almost what we did. We did about three changes a day to each of our major systems. But getting there in the government, with the way it moves, was a very slow, difficult process. Worth it, I loved doing it, and I had a lot of people who were highly motivated and really wanted to deliver well for the American people. But we had to think about how to do it. The first thought was we need to change the architecture of all of our IT systems, we need to change the way that we create software, and the cloud was the obvious first place to turn. AWS seemed like the best opportunity to us at the time. We wanted to go with the industry leader for one thing. Since we were Homeland Security, we cared a lot about security, so we wanted to make sure we worked with a cloud provider that was highly secure, highly resilient, no downtime basically, because the immigration system has to keep working. So that was how I wound up involved with AWS originally. Even that oversimplifies the story, because you don't just go to the cloud and magically everything works. To really get the impact that you want from moving to the cloud, you have to think about how do you change your processes, how do you change your culture, how do you change the organization to make sense. Because we didn't just want to go to the cloud, we wanted to use the cloud to become really fast and responsive. That's a different thing. So I learned a ton very quickly about how you can use the wonderful technology that AWS provides in order to accomplish business results, or mission results we would say in the government. That essentially is my role right now: to try to help AWS customers figure out that extension of the technology, how to use the technology to actually accomplish what they want to. I learned a lot of good lessons that I try to share.

I think AI is enabling a very different way of developing software or delivering software. When you're working in AWS, you have all these high-level services available to you to use as building blocks. This is why I say developing isn't quite right. When you're building something now, it's not necessarily a matter of writing a lot of code. Maybe you're taking off-the-shelf components from the cloud and assembling them in different ways. You're probably using some open source frameworks. There's a lot that's already done for you, and you can put it together in a way that gets the outcome that you're looking for. The principle that we started working on is how can we very quickly assemble things, try them out on users and on the public, see if they work, make modifications as we need to, use them creatively, try innovations with minimal risk, and so on. AI has a big role to play here because if you're going to deliver constantly, which was our goal, you have to find a way to test constantly for one thing. We had to set up ways that we could retest everything that we were delivering hundreds of times a day potentially, every time anybody made a change. So AI right out of the gate can help with writing those automated tests, for example, and with generating reusable code that you already know is going to work, that you've used before, with finding defects even before you have to run your tests. So it has a huge role to play, and a subtle one in many cases. It's not about writing code per se necessarily, it's about writing pieces of code, it's about finding problems in the code, it's about generating code that's secure and relatively bug free. When you combine it into this whole picture of being able to move quickly with the cloud and everything else, it has a huge business impact in that sort of indirect way. If I had had the AI tools available to me at that time, it would have helped me so much in what I was doing. It would have accelerated us a ton.

Yeah, AWS has all kinds of magic as far as I'm concerned. When I saw it as a customer, we were so focused on getting speed, and being able to use the entire tool set and way of thinking of AWS was so crucial to getting us speed that I realized everybody needs to be working in the cloud. There's no alternative really today. The second thing was Homeland Security, as I said, it's about security, and we had to work with a cloud provider that was really focused on security. AWS security is the basis of everything that everybody does at AWS. I wasn't the final authority on security, we had a CISO for Homeland Security, and I had the privilege of convincing him that we should work in the cloud. He also could see that we could have a much stronger security posture if we were working in AWS and doing things the right way, the cloud way of doing things. I came to be blown away, I guess I would say, by AWS when I started working for AWS. There were a lot of challenges in the government, a lot of things that I did that I could have done better if I had understood how AWS maintains the culture that it has internally. So I don't know if you heard the story about our 16 leadership principles, which is a crazy sounding thing. Amazon has 16 principles, and unlike most companies where their principles or their values are things on a poster and nobody remembers what they are, at AWS we actually use these leadership principles every day, and our culture has formed around those leadership principles. So sometimes when I'm talking to leaders about techniques like those that are used in software development like DevOps and other team-based techniques that are based on empowered autonomous teams, they often ask this interesting question: they say, well if the teams are autonomous and they're empowered, how do you control them? That's a very funny question if you think about it, because obviously you don't control them, that's what it means by definition that they're autonomous and empowered. But of course the question has a sense to it. How do you make sure they're working on the right things, they're doing the right things? I think Amazon has the ideal solution to that problem. The way I know that teams are doing the right thing is that I know they are following the 16 leadership principles because everybody at Amazon is constantly following those principles and thinking about them. So we have an agreement on values that we start with across all of Amazon, not just AWS. If you ask somebody who's in Singapore like Arvin, or somebody who's in China or any other place who works for Amazon, you can ask them about the leadership principles, you can ask them what's your favorite leadership principle, and you can have a good discussion, a good argument about it. So I know the team, even if they're not doing things the way I would do them, I know they're following those principles, and so the result is going to be good, it's going to be what I want. Then we have this other concept of team tenets, where the idea is that each team decides in advance how they're going to make decisions, and they write those things down as principles, and they're open for anybody to review and argue with and disagree with. I know they're going to follow those decision-making principles. So if you think about the implications of that, especially in the government where everything had to be tightly controlled and there was a rule book, you got to do things this way and this way, and then you think about how can you actually harness people's creativity and innovation and agility and make the most of them, you need some sort of solution like that Amazon solution. I found it's amazingly effective, so I'm really glad I learned about it from working for Amazon. So in answer to your question, there were some things that really impressed me about AWS before I joined AWS, but then there were all these new things I learned after I joined that impressed me even more.

Well, coming from the government, my favorite principle is Bias for Action. It's a principle that's important at Amazon. We think you can't get into analysis paralysis and take forever to make decisions. So once you've mitigated the important risks, once you've realized that you can do things without putting the company in jeopardy, basically do them. That's Bias for Action essentially. Another principle is Have Backbone, Disagree and Commit. What that means is if we're making a decision as a team and you disagree with it, you must argue. It's required. I love this one, and I love this one especially for the young employees out there. Just imagine your own company telling you have backbone, disagree and commit, disagree with your own superiors. That's right, and that's coming from the top of the company. I love it. And it's not only that you're allowed to, it's that you're required to, you're encouraged. At some point you might have to make a decision, that's the commit part. Once the decision is made, you own the decision also. So that eliminates the sort of passive aggressive thing where people later on will say yeah, I knew that was a bad idea but I didn't say. You can't, you're not allowed. The principle says don't do that. So I think that's another good example. I'll give you a third one: Learn and Be Curious. You're not fitting in at Amazon if you just say this is my job, this is it, I'm just going to do this thing. Ownership is another principle, you have to take ownership for things, you can't say it's not my job. But I think Learn and Be Curious means constantly you should be poking your nose into other things and figuring out how things work, and generating good ideas because you've learned all these new things. So those are just a few examples. The principles fit together into a way of thinking that is very Amazonian.

One point that I try to make clear is ethical is not the same as compliant. It's very easy to confuse those two things, to say give me the rules and I will make sure my company is compliant. Unfortunately, the rulebook doesn't really exist today. It's evolving. There are various rulebooks that you can use, but this is not just a matter of checking the box and making sure that you can prove that you did the things you were supposed to. It's not that kind of thing. It would be nice if it was that easy, that you have the rulebook. But for something that's emerging like generative AI, and it's changing every day, and its impact on society is not well defined yet, there isn't a rulebook. So you need a different approach to thinking ethically. Another difficulty that I see companies having, especially company executives who I'm usually dealing with, plenty of people will say you have to make sure that your AI is not biased. I fully agree, but how do I do that? It's easy to talk about those things and say you have to make sure that it's not biased, and we say that to executives all the time, but then what? What do they do with that information? I think it's important to understand that there isn't a single definition of biased. It's not even that easy. Even if you're trying your best to make sure it's not biased, what exactly are you trying to do in making it unbiased? It turns out if you study the academic literature, there are different definitions of what it means to be biased, and they're incompatible. So you actually have to make hard decisions. That was the point of my book: there aren't easy answers. You actually have to make decisions. Nobody can tell you the right way to do it. Those decisions have to be made in an attitude of caring about people, caring about customers, acting according to values, and you have to constantly be making those decisions. You can't do it once and then say okay, we're compliant, we're done.

We are so optimistic about the cloud and its power. We realize that it is applicable to social issues, social impact issues, and there are so many of them. Even where to start is a problem, there are so many things that we can be doing. We worked with the International Center for Missing and Exploited Children. It turns out that children disappear, loads in India as well. How do you find those children? So we worked with ICMEC to create a portal where images of those children could be shared. That was the first step, but then we found ways to search the web and search the dark web with image recognition to try to find any pictures we could of the missing children. In some cases we were able to find a hit and pass the information on to law enforcement to go and find the child and take care of it. This was an early application of image recognition that was so powerful in an area that was so hard to deal with. How do you find the children when they could be anywhere in the world? Actually, I was working on some projects around homelessness in American cities, which you wouldn't think could be solved by cloud technology. It's not obvious, but it can actually. Not solved, you can't solve it, this is a hard problem, but we could improve the situation a lot. The explanation of why technology is applicable: I learned a lot of things as we were doing this. It turns out that in a lot of cities, there are a lot of homeless people who are maybe sleeping rough on the streets, and there's housing available for them, and the housing is not occupied, it's empty. A huge portion of the housing that's available for them is empty. Why is it empty? It's because they have to prove that they qualify for the housing. There are different rules in different places, but often they need to show documents to prove that they qualify, and those documents might be things like a birth certificate and a social security card, all these other things that obviously they don't have if they're sleeping rough on the streets. Nonprofits will work with them and try to find those things, but it takes 18 months to do it often, and in the meantime the housing is empty. So the solution, it's not a complete solution, part of the solution is so simple: just give the people a way to scan these documents if they ever find them and store them securely online, so that when they have to produce them again, they're available. It doesn't solve how you get them in the first place, but right there it's a huge impact on getting people into the housing more quickly and solving that problem of the unoccupied housing.

Why do we even need the documents in the first place? I think in many ways India is moving towards a solution on this in advance of where the US is. Why do you have to show your social security card? Why can't we just interface with the Social Security Administration and verify that you have the card? Ways to store credentials that are full credentials that can be presented electronically rather than the paper copies, that's the real direction for the solution. From what I understand, in India with the technology stack that the government is supporting, there's the identity layer with Aadhaar, and then there are layers of credentialing on top of that, and eventually e-commerce on top of that, payment transactions. It's a beautifully thought out stack that then empowers a lot of private sector companies to use that stack to produce things that are useful for everybody. I think that's the direction that you need to go if you really want to solve some of those social impact challenges. I guess what I'm saying is the cloud is this powerful tool for doing all of these things, especially now with the new generative AI capabilities.

Well, as you say, I've learned a lot from all the different industries that I've been involved with. One thing I learned when I was in the government, they knew I was going to have to brief congressional committees, so they gave me special training on how do you answer questions coming from senators and congressmen. The first thing they taught me is never answer a question directly. They said never let us catch you answering their question directly. How do you answer them? They taught me all these different techniques, it was very funny. So I just wanted to point out that I'm going to not answer your question directly, because I don't think it's answerable in a way. My theory is we're in a very fast changing environment, and what's important is not necessarily to know right now what is going to happen in the future. None of us can really. What's really important for companies is to figure out how to be adaptive, how to change quickly, how to be agile, and how to be innovative with the technology. So it's not a question of what use cases are out there right now for generative AI, it's a question of how do you manage innovation. Leaders have to think about how they're going to make sure that their organizations are good at innovating, and then they're going to innovate with generative AI. I don't know what they're going to innovate, none of us do. But there are ways, there are techniques you can learn for being better at creating an innovation-centric company or organization. A lot of that has to do with how you manage people. I talked a little bit about managing through principles and values. Leadership can make a concerted effort to let ideas flourish and get tried out. The problem with innovation typically is people have great ideas around the organization and nothing happens with those ideas. They learn helplessness, they learn even if I have great ideas, nothing's ever going to happen. So the challenge around innovation is how do you get rid of that feeling? How do you make it possible for people who have good ideas to actually get those ideas listened to and tried out? It involves reducing risk enough that you can try out a lot of ideas. So instead of when an employee comes with a great new idea, the temptation is to say I don't know about that, have you thought about this, have you thought about that, did you talk to this person or that person. Managers have to learn not to do that, and instead to say let me help you find a way to test your idea so we'll know if it's a good idea, and then test it because the risk is so low. How do you get the risk to be low? Working in the cloud is a big part of the answer. You work in the cloud, you can set up infrastructure to test something. If it doesn't work out, you just get rid of the infrastructure and you stop paying. If you wanted to try an AI idea in the past, you would hire a bunch of PhDs in artificial intelligence, you'd give them a year to build a new model, you'd pay all sorts of things just to try out an idea. That's risky. If you're investing that much, today you take generative AI components from AWS, our high-level services, you don't need to hire the PhDs, it's already there. You try out your idea, and you see if it's going to be effective, or if you need to pivot and do it a little differently, or if it's just a bad idea and you should stop investing in it. You're going to lower the risk so much that you can encourage innovation. I want to stress that in answer to your question, because I really don't want to say how I think people are going to use generative AI, because I don't know yet. It is about learning to incentivize and encourage innovation.

It was my pleasure doing this. I really enjoyed it, and I loved your questions. Thank you so much.