Building with AI: How Twilio & AWS Are Shaping the Future of Development

0:00:05.6 Kailey Raymond: Welcome to Builders Wanted, the podcast for people shaping what's next in customer engagement, innovation, and digital transformation. Today's episode brings together two powerhouse leaders at the forefront of one of the most transformative forces in tech. AI. Joining me is Deepak Singh, Vice President of Developer Agents and Experiences at Amazon Web Services, where he helps organizations build and scale intelligent systems with world class cloud infrastructure and tooling. And I'm excited to speak with our very own Inbal Shani, Chief Product Officer and Head of R&D at Twilio, where she's driving the future of customer engagement by integrating AI into Twilio's products to make every interaction smarter, faster and more personal. These two bring a unique perspective on building scalable developer-centric platforms that power innovation at global scale. On today's show, we'll explore what it means to build with AI, the evolution of developer tools and platforms, and how leaders like Inbal and Deepak are helping customers thrive in this new era. Let's get into it.

0:01:08.9 Producer : This podcast is brought to you by Twilio, the customer engagement platform that helps businesses turn real time data into seamless personalized experiences. Engage customers on their terms across SMS, voice, email, WhatsApp and more. Power every interaction with AI so conversations feel natural, not robotic. Adapt in real time, delivering the right message on the right channel exactly when it matters. That's the power of Twilio. More than 320,000 businesses from startups to Fortune 500s trust Twilio to transform customer signals into conversations, connections and real revenue. Reimagine the way you engage with your customers. Learn more twilio.com.

0:01:57.5 Kailey Raymond: Inbal, Deepak, welcome to the show. I'm very excited to have you both with us today.

0:02:02.9 Inbal Shani: Thank you for having us.

0:02:04.5 Deepak Singh: Yeah. It's great to be here and actually great to be with Inbal on the show. It's been a while since we sat across from each other.

0:02:11.4 Kailey Raymond: Old friends.

0:02:13.0 Inbal Shani: We are.

0:02:14.4 Kailey Raymond: We'll be talking a lot about building for developers today and the intersection of AI on top of that. And so to set the stage for today's conversation, we know that you're both building platforms that developers build on top of. So Deepak, we'll start with you and then I'd love to hear Inbal's take on this one as well. But in your opinion, what does it mean to be a builder and what's the hardest part about building tools for builders?

0:02:40.7 Deepak Singh: When I think about a builder, I think about anybody who is building software. There's various definitions of builder. There's some that are very broad, but I think about it as someone who is sitting with their hands on a keyboard and building software that companies use, people use, you and I use. They're also often troubleshooting and trying to fix the software. It's not just... Most software building is actually not building it for the first time, it's actually continuing to improve an application or going and fixing things that went wrong. The fun part about building for builders is that builders are very particular, they have strong opinions because it's very important for them to work in an environment which helps them build and be in a flow, so to speak, which is why you have historically had things like the vi and Emacs battles, not because one is better than the other, Emacs is better than vi, but because builders build differently, everyone's different and they have their own particular way. So which makes it super interesting, super fascinating, and I've enjoyed it for so long.

0:03:46.1 Inbal Shani: Maybe to add to what Deepak said in terms of the definition of builder, I think we live in a world that everyone wants to be a builder. And wanting to be a builder is likely writing some sort of software or building some sort of an application or experience, which is all great, but with building software comes a lot of responsibility. And that responsibility is from when you're putting something out there, you need to carry it through. So Deepak talked about troubleshooting and continue to expand the software, but it's also getting ready for the bad days. We all know that everyone that has ever written software, maintained software, and was a product manager that was responsible on a service, you all carry a pager because the bad day is going to come and you need to be ready for that day. Even if you're building software using modern tools or AI or vibe coding and everything that is happening, but you have to understand the tool that you're using. You have to understand what is it that you're going to build, what is it going to do for you? Is it doing what it's supposed to do for you?

0:04:50.0 Inbal Shani: And building for builders is exactly that. And I think the interesting part about building for builders, you're putting a platform out there, you have some use cases in mind and you need to take into consideration that maybe on a good day you cover 60% of the use case in terms of how the builders are going to use your platform. And 40% of the time, they're going to do something that is completely different, that you never imagined. And then you need to build a platform to support it. But you also need to be understanding that the developers or the builders that are using your platform are going to do things that you never imagined. And as such being a builder for builders is really having that mindset that you're building something that is fully outside of your control, besides keeping it always functioning, so the builders are happy to continue using your platform.

0:05:36.3 Kailey Raymond: I love that. I want to hear some of those stories about the use cases that you might not have predicted coming out of the build itself. And I think embedded within both of your answers is this concept of curiosity, adaptability, flexibility that's really interesting and it's something that I'm sure we'll touch on a little bit more, especially with this seismic shift of AI that's happening in the market. I'm sure you're both feeling it very acutely every day. Inball, you lead product and R&D at Twilio, what's your focus right now when it comes to shaping Twilio's future with AI?

0:06:14.9 Inbal Shani: So let's start with there is a lot of hype and there is a lot of questions around what is your AI strategy. In Twilio, because we are some kind of a hybrid between platform as a service and software as a service, we're focusing on solving customer problem first. And it comes across with every choice of tools that we make, from AI to data to platform to cloud to services that we're using in order to build the software or the solution that we have. And it all started with understanding what are these customer problems we're trying to solve. So the question we ask ourself is not how are we monetizing AI or how are we using AI, but more in terms of how are we grounding ourselves into what are these customers' outcome that we're striving to achieve and then how are we taking that to the next level where AI tools are very applicable to create a better customer experience or a better user experience or a better customer journey. Our core products is a set of communication tools. So historically, if you're thinking about communication tools, they're very simple. You pick up the phone, you start a phone call, there is a person on the other end and here you go.

0:07:23.8 Inbal Shani: Now, in the world of AI, there is more AI agents that are embedding the support experience. So how are you building that experience in a way that it feels seamless for the customers and it's much more enjoyable and it's not the legacy, it's an IVR, click one and then what is your problem? State it again. So that's one experience in terms of where we're thinking about AI. But I think one more thing to add into that is AI is only as good as your data. And we are often talking about AI, but not so much on the data. When we're thinking about building AI experience, we start first with being able to give that contextual data to be able to build these AI experiences. And that comes across in terms of how are we building our platform, what is a cross channel orchestration, what is the customer journey and everything that we're building as part of the Twilio platform.

0:08:19.1 Kailey Raymond: That makes sense. I mean what you're talking about to start is really just customer centricity as the basis of the roadmap. And it's interesting that kind of baked into this idea of contextual data too I think is this concept of speed which I think we see in the market acutely, where relevance has a lot to do with the pace of your understanding of customers. So that kind of idea of real time is almost inherently baked into that concept of real time and how good is your data? Deepak, at AWS, you've been on the bleeding edge of cloud and the AI evolution. What's your current focus and how does it support builders across industries?

0:08:54.4 Deepak Singh: Well, right now my focus is giving software developers everywhere, whether it's inside Amazon or outside Amazon, tools to help them build software and using AI very specifically, which the fun part of it is every two months everything you do feels like it's obsolete because there's a new model, there's some new science coming out. So you talked about customer centricity earlier. I think the fun part about what we're doing right now is I think people are still just beginning to realize what they can do with these tools and what they can do varies from Inbal talked about hype earlier is understanding what their strengths are and what they can't do and knowing very well that those will change very, very quickly because what you can do now, you couldn't do six months ago or nine months ago. And as a product person, it's both fascinating because customer is also trying to figure out what they want to accomplish because they're also understanding the tools at about the same time as you are. I'll give you an example, probably put some bow on it. There's a very senior engineer at Amazon who used to be an AI skeptic and he came, sent a note to some folks saying after using one of the AI tools for a while, he said I am 4.91 times, very specific number, more productive than I used to be before.

0:10:10.9 Deepak Singh: And then walked through why. And having people have that aha moment where they understand how they themselves can maximize these tools is worth all the challenges of working with the strengths and weaknesses of LLMs.

0:10:27.7 Kailey Raymond: I would love to learn how I can be 4.9x more productive on a daily basis. So connect me with this engineer after the fact. I mean, listen, we're seeing this wave of change with how folks are starting to build across teams. I think that you both are kind of touching upon this. So I'd love to hear this from both of you, from your perspectives, what's the most exciting or urgent shift that you see happening in the developer ecosystem right now? Inbal, why don't we kick off with you?

0:10:56.6 Inbal Shani: I think Deepak touched on it before. It's all about speed and it's ease of use. What we've done throughout the years, we've asked the developers to become these superheroes. They get to do more things in their day than any other person in the organization, which basically when you boil it down, they get two hours a day to write code. So if you think about an engineer working 12 hours a day, two of that, they get to write code. So what they're looking for is something that will help them get more productive. So if maybe the 4.91, maybe it's the times 10, maybe it's the time 3. Each one of them depends on the type of work they do. But it's all about becoming more efficient. And that's the trend we see. We see that developers are adopting AI tools not to replace them, but to make them more productive. So when they are in that flow, when they are writing code, they can solve bigger problems, they can implement better system design, they can think about all the issues that might occur, they can think about better use cases versus just spending time on kind of that boilerplate code.

0:12:02.8 Inbal Shani: That doesn't necessarily make them more effective, doesn't necessarily make the solution better. And this is where they can use AI tools and we see more and more kind of in that developer mindset is I have no time, make me productive from the get go. And it comes across from how fast can I onboard on a tool, how fast can I use it? Can I understand what it is doing for me, what is the learning curve and how we're shortening that learning curve and how fast can I get really return on my investment in spending time on learning that tools? And that's the biggest shift we see in kind of the developer mindset when it comes to adopting these tools, even with AI skeptics.

0:12:42.5 Kailey Raymond: Time to value efficiency. Anything you're seeing, Deepak?

0:12:46.3 Deepak Singh: Yeah. AI tools for software development and building are a journey. You used to have... On one side, you have the skeptics. On the other side, you have the people who think you press a button and unicorns and rainbows come out from the other end. I think over time, a lot of folks learn how to use these tools in a way that fits them. I actually think one of the biggest changes that has happened in the last six to nine months is agents. A year or two ago, all you had was chatbots and auto completion. That worked up to a point, but I don't think it was going to change the world. I don't think anyone was in a place where they felt that they were going to completely change the way they build. But I think what started happening now and this boils down to what people do, how they operate, which is why I think I see senior engineers actually gravitating towards these tools a lot these days, because it helps them think and get stuff done while continuing to do what they do best, which is help other people solve problems.

0:13:41.0 Deepak Singh: And I'll give you an example of a couple of trends I've seen in how people work. And this is because they're understanding these tools and their strengths and weaknesses. There was a time that people would ask a function, write it down, copy and paste it from their chat window, but as agents start writing tools and you are driving, remember, the human is the one driving these agents to where they want them to be. What I'm starting to see people do as they trust them a little more is say, you know what, I'm going to generate all this code upfront. I'm not going to go function by function, and then I'm going to spend my time refactoring. And your refactoring might also be done by conversing with the agent, et cetera. But the way you are building has suddenly changed because you have built a level of trust, because you understand how to prompt, you understand where you need to look. Inbal mentioned data. I say context listing and data is part of your context. What data are you providing the agent to build? And it's been fascinating to see this shift both inside Amazon, where I see our developers, my developers, with friends, family, anyone writing code. It's been a really fascinating journey, and I would say the last six months in particular. So we live in interesting times.

0:14:48.9 Kailey Raymond: Yeah, we sure do. The pace of change is unbelievable to sit and watch. And to all of your points, I think this concept of curiosity generally keeps coming back is, if you aren't able to dig into something and ask the right questions or know what a software can or can't do or a platform can or can't do, like you're not in a great place. Inbal, this one's for you. We have customers across industries, so I'm wondering if you're noticing any nuances in industries or across sizes of companies. We talked to startups, we talk to enterprises. What are our customers asking right now when it comes to AI driven engagement?

0:15:26.2 Inbal Shani: I think the interesting is that the trends are the same, the size and scale is different, but if I divide it to kind of three buckets, I would say with the AI skepticism, all our customers are asking for trust. Why should I trust you with your data? How do I know that your platform is doing what it's supposed to do? I'm giving you my treasures. How do you make sure that you're protecting them for me? And then especially if we're putting now AI in kind of areas that are critical for my business, how do you make sure that this AI is reliable, it is responsible, it is secure and safe. I think the second part, which is what we talked before, is simplicity because all companies want to get started fast. If there is a long learning curve around these AI experiences or any experience, they will tend to pull back because they want to get value, they want to realize value out of your solution very fast. So you need to make onboarding simple. You need to make even finding documents or how to get started or writing functions in a very fast and easy way.

0:16:28.8 Inbal Shani: And I think the third one is they all want that wow moment, right? They all are looking for that smarter experience. In terms of here is where AI or an AI tool can do something that maybe I couldn't achieve before. Maybe it creates like a different level of experience. You're taking few steps forward instead of multiple steps to get to the same place. And these are kind of the three trends that we see mainly from our customers. And again, it's startups, it's enterprise, it's our ISVs, it's all across the industries. And if you think through these lens of these three buckets, then it helps you take the right decision in terms of which tools you're using and which AI experiences you're building.

0:17:07.3 Kailey Raymond: That's a nice little framework. So trusted, simple, smart are kind of the three themes that you're pulling out from our customers across their sizes and industries. Deepak, we've talked a little bit about this idea of pace and the pace of change and I would love to hear from you about how AWS developers are approaching AI today versus a year ago. What new patterns were you seeing across cloud usage, machine learning? Walk me through that.

0:17:34.7 Deepak Singh: Yeah, I liken the use of generative AI to go into a swimming pool complex. About a year and a half ago, most of the customers we talked to were in the children's pool learning how to flap their arms. About a year ago, I would say, generalizing, a bunch of them ended up in the regular pool, started making their way towards the deep end. There were some who went all yolo. We'd go to the Olympic pool, like deep. We'll figure it out. But I think there's some common things to the ones who've been successful. Curiosity is almost constant, which is the customers will be more successful. And to Inbal's point, it's not just the startups or the small companies, I've seen this with companies of every size. The customers who have been most successful are the ones who enable curiosity because this is a world where there's so much changing that if you say, oh, we're going to have this, your first instinct is to set up this big governance thing. By the time you're done setting everything up, the world has moved on and then you have to set up another one. So figuring out the sort of boundaries under which people can be curious and enabling smart people to do smart things is where I think the people who are successful have been the most successful and others have realized sometimes the hard way, that that is the way to go.

0:18:50.8 Deepak Singh: And I think a lot of from ourselves, as vendors and providers and as cloud people, the thing we can do is help there. Very specifically for our customers I think some of the trends are exactly what I've talked about. It's once you build trust, what do you build. People start off with a chatbot or a simple app here or there, but we have customers who push things into production and actually the fun part is they've often sometimes done it without actually typing a line of code. It's all being, for lack of a better word, white coded or prompted is probably the more current technical term. It was fascinating to see and this is a large company which you would, like, really, they did that? But they did because they got comfortable over time on what works, what doesn't work.

0:19:33.0 Deepak Singh: And the other part that I've seen is again, I think Inbal touched on it, it's not just the things that sound really cool, it's upgrading old versions of Java, refactoring applications, automating things that came in the way, taking this JavaScript, this SDK that you wanted to upgrade forever and figuring out how to upgrade it and getting out of the way of people. Sometimes it's the mundane stuff that actually has the most impact to a company, but the time to wow is really important because you can't get people excited. In 30 seconds, they move on, which is one of the nice things about AI, I think that happens a lot.

0:20:06.9 Deepak Singh: So it is a very interesting world because there are more and more people who you think would be conservative and they're putting guardrails, not gates, at this point of time. And I think it is starting to make an impact on how they operate, which is going to be very interesting to see how that plays out over the next few years, but it's across industries. I don't think there's any industry that I can point to that says they're not jumping in and trying this out.

0:20:31.0 Kailey Raymond: It's interesting, often on this show, what it boils down to is we realize that a lot of it comes down to the boring, unsexy stuff that actually makes the biggest impact at a lot of companies. But I really love this concept that you baked into, what you were talking about, which is enable curiosity. I think that that is a real beautiful phrase to kind of take away from this conversation. And what we're talking about is, organizations and teams kind of fundamentally adjusting their processes and culture to take care of this new AI powered, real time software development advantage. We've talked a little bit about, you said guardrails. I'm wondering, and this one can go to both of you, I'll start Deepak with you. How can they manage the risks? Especially when it comes to the pace of innovation, security. There's all these fast changes that are happening and sometimes these decisions are irreversible. They might be permanent. So walk me through your ideas there of risk tolerance.

0:21:31.8 Deepak Singh: Yeah, actually one way to do it is to try and do things that are two way doors and not one way doors because it makes life a lot easier. What we found is, and I'm talking from a big company perspective, if you're a small company, the same person is making all the decisions, so it's a little bit easier. But in larger companies, sitting down with your AppSec people very early on and figuring out the rules of the road is really important. And what I found is most application security teams and most CISOs, they want to enable the business. They see the value of, so they know that they can't come in the way, but they're also making sure that they're setting some the rules of the road. Don't do X, don't do Y. I'm making this up. But if I'm a financial services company, don't put my risk management software into this yet. But other things, may be just fine from a security company as well. There are certain things that you will be very comfortable doing and others like, let's talk about it. And I think when you set up those frameworks earlier, that's good. The other part is the tools also help you.

0:22:29.1 Deepak Singh: I mean you can add guardrails. Like if you're writing a chatbot, you want to make sure it doesn't swear at you. Adding guardrails is not that hard. It's software like some of the stuff that we have you can add in reference trackers which make sure that you're not putting any copy left code in there so it flags that for you. So the tooling also helps. It's gotten more mature. Observability has gotten a lot better. But I think like everything else, sitting down and having a discussion early and just setting down the rules helps a lot because you don't want to do it all the time. And most people think that application security folks are there and the CISO is there to show you the hand. And in my experience, they want to enable the business. They just want to do it in a way that doesn't become a one-way door, that we can't recover from. And the people who do that early and in a lightweight way tend to be the most successful in my mind. And people are getting comfortable with AI also, compared to where they were a year or so ago. Your thoughts?

0:23:27.0 Kailey Raymond: Anything to add there Inbal?

0:23:29.1 Inbal Shani: I think it's defining the framework for experimentation because you want experimentation to happen all the time and you want them to happen fast. But it's all about the differentiation between this is an experimentation to this is going to production and the customer impact. You need to have these checkpoints along the way. You can experiment with everything and anything as long as it's within the guardrails that were defined. But when it comes time to make it into a customer facing experience or when it comes time to take it into production, this is where you need to start having these kind of one way door, two way door decision. It's what is the impact? What are the risks? And start managing the risks against that. But you want to encourage innovation to happen. And then the other part is there is decision that needs to be centralized and then decisions that should be decentralized. But for example, what type of vendors you're working with? Are these tools you should rely on, you should bring them in or not? And then having the AppSec team or the data privacy team kind of making sure that they are validating your choices even in the experimentation phase will help you move much faster.

0:24:33.8 Inbal Shani: And then making sure again that you're not taking these one way door decision that you're stuck with and now to take it into production, it's a full rewrite or maybe even the experiments is a risk to the business. So managing the risk and the speed of experimentation and knowing when to centralize and decentralize.

0:24:51.0 Kailey Raymond: I really like that thought around centralization versus distribution of decisions. It could get really out of hand really quickly if you're just saying to every team, go use it, without any guardrails in place. And conversely, you want to enable that curiosity and that innovation so you need to give them a starting place.

[music]

0:25:10.2 Producer : Great customer experiences aren't magic, they're built, and Twilio is the platform that helps you build them. Every customer action, browsing your site, opening an email, reaching out to support triggers instant AI-powered personalized engagement across SMS, voice, email, chat and more. No delays, no guesswork, just the right message at the right moment. From automated messaging to seamless authentication to hyper personalized customer journeys, Twilio's customer engagement platform powers millions of interactions daily, helping businesses drive loyalty, optimize marketing spend and create experiences that people remember. Twilio is the ultimate toolbox for customer engagement, ready to build experiences that matter, visit twilio.com.

0:26:01.7 Kailey Raymond: We've talked about this a little bit, about the hype. I think you've both used the word so far. There's no denying it. There's also a lot of pressure around AI, so this one can go to both of you as well. I'd love to hear what you think people underestimate when building AI products at scale. Inbal, why don't you start?

0:26:21.5 Inbal Shani: AI has a learning curve. Getting started is easy, but if you want to make the most out of your AI tools, you need to truly understand what they can and cannot do, what's their weakness, and what's their best use for, so understanding that AI has a learning curve, I think this is an area that people tend to underestimate because it's so easy to get started. I think the second part is there's a cost for iteration. So when you're using AI, you need to understand that using a model, training a model, having the data, having a feedback loop, all of that is not a magical thing that just happens by itself. It requires investment and you need to be serious about it. I think these are the two things I would say, and likely the third one is that real world problems are very complex. They're not as fast and easy like I'm going to a ChatGPT and asking it the question, it gives me an answer. So when looking to take AI into production, you need to understand what is the complexity of the problem you're trying to solve and how to deploy the right AI solution to really solve that problem versus swinging by.

0:27:26.5 Kailey Raymond: I like it. And Deepak, I saw you smile during that answer, so I'm sure you have something to add there.

0:27:32.0 Deepak Singh: It's been an interesting journey because one of the things we've consciously done is build our software using the tools that we built. Because the best way to learn how to use it is to use it. We've learned a few lessons. The first is engineers have to understand science more than they used to. In the past, especially in the classic machine learning, deep learning days, it would be very easy to take something and say, oh, this is a science problem. I'm going to throw it across the Jupyter Notebook to the science team so that they can do what they do and come back. It's not that, it doesn't work that way. The science is the product, right? When you're building an agent, it's a collaboration between science teams and engineers. But engineers have to think scientifically as well, and scientists have to think like engineers a little bit. So I think the people who figure that out faster and more effectively are going to be more effective. I think that's going to be a big part of how building is going to evolve. Second part is some things never change. And the way you build software, the things you have to be careful of, doing testing, et cetera, it's the same.

0:28:29.0 Deepak Singh: You might think it isn't, but it is. But some of the techniques and approaches you take, change, especially as you move more towards agents. And now you think about APIs are still deterministic. Your API has a response. AI doesn't. I mean stochastic by definition. GenAI is stochastic. And actually for some engineers, that's hard to wrap their heads around. So then figuring out how do you do evaluation, how do you test, et cetera, like this, it changes the practice of engineering and seeing teams figure out what stays and what changes have been quite fascinating over the last year or so. So in addition to everything that Inbal added, that's the part that I will bring in from just the experience that I have seen with teams and it's been a lot of fun actually, just seeing them adjust. Also half the things you learn you know you'll have to relearn. For example, Claude 4 has interleaving, interleaved thinking, which is the new way of doing things that suddenly half the folks on the team are trying to figure out what can I do with that, three months after the last model. So it’s fun.

0:29:38.5 Kailey Raymond: This is perfect. You set me up for this question that I have for Inbal very, very well, which is Twilio has long been known for their developer friendly APIs and AI changes the game a little bit there as Deepak just kind of mentioned and the way that engineers have to think about this and we've talked about this concept of simplicity, trusted, simple, smart. How do you think about making complex tools feel accessible and trustworthy?

0:30:05.7 Inbal Shani: Yeah, well we're lucky. In Twilio, simplicity is our DNA right, when we built our first API ever, it was to abstract simplicity. So our DNA is all about how are we making developer's life easier. And now, we need to think that it might be a human developer or it might be an AI agent that is interacting with the APIs or they're building the next experience on behalf of something else, someone else, maybe it's a different AI agent. So when we're designing products right now, when we're thinking about services, when we're thinking about the new platform we're building, we're taking that into consideration that things are not necessarily going to be deterministic, things are not going to necessarily going to behave the way we expect that. And it's coming across... You know, we're designing a new console and a lot of the questions we're asking ourselves in the new console, okay, is there going to be a dashboard, is it going to be a human that will look into these dashboards or is it going to be an AI agent that will need to report back to headquarters, like here's what's happening with your Twilio implementation. So how are you thinking through that?

0:31:09.8 Inbal Shani: That now the engagement is not necessarily that you have a human in the loop. Maybe the human is kind of disconnected or there is a layer of separation that is called an AI agent. If it's around your identity verification, if it's around building these engagement, if it's support use case, even if it's about scheduling an appointment. And these are the guiding principles that we deploy when we're thinking about these new experiences that Twilio is building for the customers. One of the interesting things that we're kind of debating, we are part of a regulated industry. This means in order to send an SMS, do you need to register as a business? What if now there is an AI agent that is representing that business? How do you think about verifying that that AI agent is really representing the business that they say it is and not something that is maybe a fraud. So it opens up like a brand new thinking. When you're thinking about the world that we know it, that identity is unique and an API gives you a deterministic response to now a world that everything, you can expect everything and anything to happen. And now you need to design for that.

0:32:12.0 Kailey Raymond: How do you design for that? That is like unbelievable. I mean you're talking about this idea of abstracting complexity which goes back to this idea of simple. In a world in which the possibilities are endless and there's so many decisions to make, obviously that's really hard. So maybe Deepak can help us answer this, which is like, how do you think about balancing all of these things? Performance, responsibility, customer needs. Like what's some of the approach when you think about these trade offs?

0:32:38.8 Deepak Singh: There's two things that come to mind. The first one is listen to your customers. In the end, the biggest mistake I can see, and I've seen it happen in the AI world is because it is intellectually stimulating. It is very easy for a dev team or a product team to say this is the best way of doing something and we are going to follow this to wherever we want to instead of working with customers to understand do you even care? I think that's very important. The second one is somebody once told me that you have to unconstrain your thinking. I think with AI, it's very important to take constraints away. The traditional world has constraints because AI works differently. Your brain is not quite instinctively, no that's not possible. Then you realize it actually might be possible. So taking away some of those constraints in your own thinking, both individually and as a group is really, really important. So if you combine listening to customers and unconstraining your thinking, you can accomplish a lot in relation to every think that Inbal said. And there's some examples like that you realize as you go along.

0:33:45.0 Deepak Singh: The first one is you're now coding more in natural language than in a programming language. And interesting output of this, this is what comes if you listen to customers is by the way most people are very comfortable typing in a programming language because there is no first language, second language, doesn't matter what your native language is, but when you're doing it in natural language, suddenly that becomes very important. So supporting multiple languages and not just English is significantly important. Understanding what a piece of code does in the past, I remember once I got a piece of code where all the comments were in Polish. I had to find the person who had written the code to... And it was in a language I didn't know very well. And that was not fun. Now, it's trivial. You just ask your AI assistant to tell you what it is. It'll tell you in whatever your language is, Telugu or Hebrew or whatever. So I think that is simple, but it actually becomes, it becomes important and it's actually very useful. So I think that's an example of the kind of thing that instinctively you don't think about. But if you talk to your customers and you understand what's possible, it's like, yeah, this is going to make a difference.

0:34:49.1 Kailey Raymond: I love that example. That is so, so funny. And one of the things that you said, listening to your customers, I always think about that. That's obviously like the ultimate kind of feedback loop to know that you're headed in the right direction. That's why you build right and so Inbal, I'm wondering if you have one of those moments this like this is working moments from the customer side, from inside of Twilio that might have showed you that, this whole AI thing that we're investing in is actually paying off.

0:35:18.7 Inbal Shani: I remember early on in my career I was tasked to build different virtual agents for support use cases mainly. Sometimes for consumers, sometimes for enterprise. And we all have... Anyone that has been in my shoes and needed to build these different virtual agents, the chatbots still have the battle scars to show for every workflow that you need to think through and then all the if else branches that you need to think through when it comes to having this engagement with a chatbot. And as consumers, we truly like to dislike these experiences because they are painful. They are far from having the best customer experience. So going back to today and what AI really enables us to do is take these virtual agents into a much more natural conversational world. One of the products we have recently relaunched was our Conversational Relay and we've announced it in our Signal Conference a month back. The interesting part is, first of all, we have seen crazy adoption. The moment we have launched that product, convincing customers to use it was not an issue. They will all help us get off these virtual agent that is legacy that requires us to click a button or build these workflows.

0:36:30.9 Inbal Shani: So how can we build that into our voice customer support? One of these companies is a healthcare company and they were able to be one of our first early adopters of that experience and build their own virtual agents into their voice system. And the interesting part, they have seen like the customer satisfaction score going up. Historically, every time you put a virtual agent, your customer satisfactory would go down, because I would like to talk to a human that can understand what I'm trying to say. But now, they put these AI agent as part of their voice flow and suddenly, their satisfaction went up. Because I don't need to wait in line to talk to a human. It understands what is it that I'm trying to achieve and in a high percentage of the time, it will solve my issue. And maybe it's even a sensitive issue that I don't really feel comfortable talking to a human, especially when it's a healthcare. Maybe I just want to refill, maybe I want to leave a message to my doctor, and I don't want that human interaction. And that is a moment that as a product leader, as an R&D leader, you are saying, wow, this really works because you see the customer satisfaction.

0:37:34.4 Inbal Shani: And I think the interesting part is after they have launched this experience, they went around and they talked to other customers about how amazing this experience. So when your customers are becoming your strongest advocate, that's the moment you know, yeah, we got it, we solved the problem for the customers and that was an amazing experience.

0:37:52.6 Kailey Raymond: I love that. Yeah, some of those customer stories are so powerful from the nonprofit customers too. Oh my gosh, some of those were excellent. And Deepak, I want to hear from you too. Inbal mentioned this earlier that about 60% of the things that you can think of in the use cases are actually leveraged. But there's another 40% that means of use cases that maybe you never even thought of. Maybe this is one of those standout examples that you come to your mind about teams using AI in creative and unexpected ways.

0:38:23.7 Deepak Singh: Yeah. I'll give you two examples. One is actually not people writing software, but it's an example of what happens when you allow people to be curious and create something. About a year and a half ago, somebody in my team created an app to teach prompt engineering. But he did a couple of smart things. One, he got approval from AppSec very early on on what kind of data you could put in it it. And second, he made it viral. He allowed you to share the output. So the thing it did, it eventually got there after a few iterations was you typed in what you wanted to build and create an app for you, a very simple app. And turns out, next thing you know, 100,000 people inside the company were using it. My favorite example is a bunch of folks in our data center. There are a bunch of data center techs who took care of the air conditioning and they built an app for themselves where they uploaded all their... I don't know if you ever had to deal with air conditioning and HVAC errors in your home air conditioning systems, may get an ERR43 and you have no idea what it is.

0:39:21.1 Deepak Singh: Just imagine that with a data center. So they uploaded all their documentation and created an app where they could just enter the error, it would tell them what it was and then they could troubleshoot it. Nobody would build that app for them. It's not worth hiring a software team to do it. And they did it by themselves. So it's like when the cost of building an app and maintaining it gets low enough, people get very creative and solve their own problems. So that's one. The other one and I'm pretty sure when that app was built by the guy on my team, he never expected somebody to do that. I'll just read out part of a Slack I got. I didn't even know our agent could do this. He says, I just saw the agent perform a manual task, then stop and write a Python script to automate the work I gave it. It started by analyzing the required changes in the JSON Markdown files. And after manually updating a few files, it recognized that this was a repetitive pattern and wrote itself a script to automate the rest of it. This is not the human asking it to do it. The agent decided to do it himself. I didn't even know we could do that.

0:40:22.4 Deepak Singh: So unexpected. Now we are thinking, okay, how do we add new capabilities which make it easier and so on and so forth. But these sort of emerging behaviors in AI lead to interesting product ideas. And as I said, these are fun times. And I'll keep saying that because they are.

0:40:38.9 Kailey Raymond: That's unbelievable. The AI is recommending the next step in building something on your behalf already. That's awesome. I love it. Those are kind of like individual examples. I love both of those. You both support huge developer ecosystems. I'm wondering at like maybe a more aggregate level recent signal. Maybe it's from Usage Data Feedback, maybe your communities that got you really excited about where things are headed. Inbal, we'll start with you.

0:41:07.9 Inbal Shani: Yeah. So if you ever build a platform that developers use, you know that the moment you put something out there, they have a list of here's all our complaints and ask for improvements and here's all the things that doesn't work and go fix that for you. So that's the usual signal you're getting when you're putting something out there. I think for this specific trend that we're seeing now when we're building some of these new experiences and we're building different things that operates differently is that you get feedback from developers into what's next. And it's not in terms of, hey, fix these issues that I have, but in terms of this is amazing, what can I get more like, how can I build faster? How can I get a better experience? Are you going to support multiple models? Are you going to open it across all the channels? Are you going to take it to the next level that maybe I don't need to work with that SaaS application or that SaaS application? So you're getting not just a lot of positive feedback, but you're getting a lot of what's next type of questions and then a bunch of recommendations because they already have different ideas that maybe you didn't think about.

0:42:10.3 Inbal Shani: And getting that feedback loop so quickly coming in, it also shows you that one, they were able to adopt it faster and second, they have already ideas in terms of here's what else you can do for us. And that's a great excitement that is happening.

0:42:24.8 Kailey Raymond: They're helping to build the next roadmap for you by giving you all those suggestions. I love it.

0:42:30.4 Inbal Shani: All the time.

0:42:32.3 Kailey Raymond: Deepak, anything to add there?

0:42:33.8 Deepak Singh: Yeah, I'll actually use an example from healthcare. I think the kinds of things that are being enabled that you know, make you want to do this. Obviously, the cool stuff like the ones I just mentioned are part of it. But you have folks like Genentech, who are doing drug discovery and enabling their own scientists. Like the search space in the human body is pretty much infinite, number of potential biomarkers, et cetera. So like they're giving people the ability to build an agent that does biomarker validation. Scientists don't have to run thousands of experiments just to figure out what to do. They're seeing that happen and make it like it's part of how they work now. And examples like this are going from once in a while to hearing a new one every other week. So I think those are the kinds of things that that make you super happy because that's also very meaningful and has an impact on healthcare, so, which happens to be the one I lived in before I came to Amazon, so makes me even more excited about things like that.

0:43:33.9 Kailey Raymond: I love it. Those are the examples where I feel like the skepticism, the hype, all of that, that grounds you and gets you really excited about the potentials of AI in the future. I think it's something personal that everybody can relate to. That's awesome. I'm going to pivot. We're going to have some fun. I mean, we've been having fun this whole time, honestly, but we're going to have some, like, lightning round questions for both of you. So are you ready? These are both for both of you. Okay.

0:44:03.1 Deepak Singh: Bring it on.

0:44:03.9 Kailey Raymond: Okay. Beautiful. Whoever you know chimes in first, we can do this, really fun...

0:44:08.3 Inbal Shani: Cool.

0:44:08.9 Kailey Raymond: So, yeah, it's like a buzzer, you know? What's one personality trait that every builder needs to thrive in the AI era?

0:44:17.7 Deepak Singh: Curiosity.

0:44:19.9 Inbal Shani: Fail fast and move on.

0:44:21.7 Kailey Raymond: Fail fast and move on. Curiosity. I love it. Favorite product or platform that you've worked on this year?

0:44:28.9 Inbal Shani: Our traffic intelligence, like the ability to think through how to protect and secure communication. Also improve communication using different AI models really was wow. And I would add the second one before Deepak jumps in, our trusted, simple and smart platform.

[laughter]

0:44:47.3 Deepak Singh: And for me, it's Amazon Q CLI, the example I gave where it felt like it had this emerging behavior was from the Q CLI. It's what I use every day. It's fun to use. Also, customers love it. Can't complain.

0:45:00.3 Kailey Raymond: What's a tech buzzword around AI that you'd love to retire?

0:45:05.7 Inbal Shani: How much time do we have? There is a long list.

0:45:08.2 Kailey Raymond: I'm ready to hear it.

0:45:11.0 Inbal Shani: AI will replace developers.

0:45:14.2 Deepak Singh: That one Inbal took... I will say democratization, because I actually feel like these tools, they make people effective regardless of their skill set, but they make the skilled people even more effective. So kind of the other way of saying what Inbal said.

0:45:26.8 Kailey Raymond: You're talking about marketers, so all of these buzzwords were invented by people like me. So it's really actually, let's kill them. Let's get rid of these. You're right, they're not great. This is the last question for both of you. I'd love for y'all to provide an example of a company that you might admire with how they're using and building with AI.

0:45:50.9 Deepak Singh: I'm going to cheat. I'm going to mention a company that has enabled AI. It's a company I've admired for a very long time. And that's Nvidia. And the reason I picked them is, I remember 25 years ago when they started getting into scientific computing and high performance computing, the world I lived in at the time with computing libraries at CUDA, 25 years later, none of this would have happened if they hadn't gone towards GPUs and invented a programming model that all of us are using for building these models today. So I don't think anybody at that time appreciated. In fact for the longest time people thought there were like 55 people in the world who cared and they were all in the same lab at Harvard, but it changed the world. So I'll pick them even though it's more of an enabler than somebody using.

0:46:42.6 Kailey Raymond: Pretty important to what's happening today, I'd say. Inbal?

0:46:44.8 Inbal Shani: I'll choose something that is close to my heart, so you can hear that English is not my first language. And as someone that has lived in different countries, some of them speak English, some of them not so much. Multi-language translation has always been a challenge. And as someone has built virtual agents throughout her career, you know that the moment you need to do a language to language translation, you always get like a mixture of salad of words that most time doesn't make sense. So the company that I truly appreciate is ElevenLabs. They have put a lot of emphasis in terms of the multi-language translation and trying to make it as real time as possible. And I think one of the use case we had a Twilio hackathon and one of the use case that was demo using ElevenLabs production was actually a call center in a refugee center where you have people from all across the world supported by different people that speak different language. And what happens when language is not a barrier anymore? And you know that moment gives you goosebumps because it's like hey, this is actually AI for good.

0:47:46.5 Inbal Shani: It's not about commercial usage, it's not about selling you more things, it's about helping people. And then you see the true value of how these technology can change the world for the best and remove a lot of these barriers that we have tried to remove for many, many years.

0:48:02.7 Kailey Raymond: Well, Inbal, Deepak, thank you so much for being here. This has been a really fun conversation, really appreciate the time.

0:48:10.5 Deepak Singh: No, thank you.

0:48:11.2 Inbal Shani: Thank you.

[music]