Working Backwards: AWS’s Customer-Centric Approach to Agentic AI

Ever wonder how the giants of tech build AI that truly understands your needs?

Get an in-depth look into innovation that genuinely serves user needs in this dynamic session from the Six Five Summit: AI Unleashed!  Rory Richardson, Director of Next Generation Developer Experience, Gen AI at Amazon Web Services (AWS) joins host David Nicholson  as one of our Enterprise AI Speakers to share a hyperscaler’s approach to integrating customer feedback into the development process.

Key takeaways include:

🔹Customer Feedback as the AI Compass: Explore how AWS profoundly integrates direct customer feedback into every stage of its AI development process, ensuring solutions are built "working backwards" from user needs.

🔹Crafting User-Centric Agentic AI: Delve into the critical challenges and careful considerations involved in creating Agentic AI that is intuitively user-centric, enhancing experiences rather than complicating them.

🔹AWS's Ethical AI & Future Directions: Understand AWS's robust approach to ethical considerations in AI development and deployment, alongside a forward-looking view of their future AI capabilities and comprehensive developer support.

🔹Innovation for Next-Gen Developer Experience: Gain insights into how AWS is continuously innovating to empower developers with the tools and environment necessary to build sophisticated and responsible AI applications.

Learn more at Amazon Web Services.

Watch the full video at Six Five Media, and be sure to subscribe to our YouTube channel so you never miss an episode.

Or listen to the audio here:

David Nicholson: Welcome to The Six Five Summit: AI Unleashed. I'm Dave Nicholson, and for this Enterprise AI Spotlight, I'm joined by Rory Richardson, director of Next Generation Developer Experience and GenAI at Amazon Web Services, which we all know by the power acronym AWS. We'll be covering AWS's customer-centric approach to agentic AI. Welcome, Rory.

‍

Rory Richardson: Thanks for having me.

‍

David Nicholson: It's good to meet you. Thanks for being here. Let's dive right into this. AWS has been involved with AI for quite a while. Can you kind of walk us through the history of how AWS got to where AWS is now in AI?

‍

Rory Richardson: We started this journey well before the rise of generative AI. We started incorporating AI into the fabric of our services several years ago, so that things like optimizations or performance analysis, we've been using AI tools in the fabric of AWS for quite a few years now. The generative AI journey, though, is fairly new to just about everybody, which was two years ago I think we went GA with CodeWhisperer, which I think was our first service that natively integrated generative AI into the fabric of the service back in April of '23.

‍

David Nicholson: Would you agree that we're sort of at the dawn of the age of agentic AI? And if so, how about if you give us your definition of agentic AI?

‍

Rory Richardson: Oh, I'm glad you said it was my definition, not AWS's definition. This is a definition that keeps evolving as the characteristics of agents continue to take on more properties or different properties. Personally, I like to land on the definition of agent with the three A's, but last week I added a fourth A. So I'd like to think of agents that is different than an application because it can form tasks autonomously. It can do things asynchronously, and it can do so with agency, like in a pride and prejudice sense, being able to make some decisions on its own and have agency. But then I started to think about the rise of the super agents, which is what I see is coming next. And so I added a fourth word, atomic. Because now we're thinking about agents being more modular because we're starting to see the agent to agent communication protocols standardizing. Once that happens, we're going to be able to build more and more complex things, just like with Legos, by combining different agents to work together.

‍

So when I'm thinking about what is the scale or the scope of any given agent, I'm biased in that I tend to think of it in terms of a microservice architecture and create an atomic mindset or framework for the agents so that we can combine them in a gazillion different ways to form super agents. Which the term super agents, think of it almost like a persona-based way to get something done. For instance, think about an SRE, a site resiliency engineer, a site resiliency engineer has, as a human, a lot of different things that they can do. And these would each be separate agents sort of underneath the covers. One thing that we learned recently is maybe we shouldn't spend a lot of time trying to teach people different agents to go to. If I teach you, if you want to do a unit test, review that slash test. Wouldn't it be cooler if you could just ask, hey, can you form the unit test on this corpus of code? And so we don't have to adopt an artificial language or another abstraction layer, but we can really get to natural communication just like you were talking to a human.

‍

David Nicholson: When you use the term super agent, you're talking about an aggregation of agents as opposed to implying that there's an agent that is orchestrating the other agents?

‍

Rory Richardson: Oh, actually the latter. I think about it in terms of being a concierge. So when you go to a concierge at a really nice hotel, they handle everything for you. I don't know if you've had this experience, but I've stayed at a really nice hotel in China, and this concierge was amazing. You just say, hey, I'd like to have dinner at a nice restaurant. They would pick out the restaurant. They would order the driver. They would give you a translated menu. Everything would automatically be handled based on the intent that they understood of the experience that you wanted to have. So having an abstraction to handle the orchestration, picking what agents are best for whatever task, and then aggregating that result, well, it's more human, it's more natural in going from intent to the thing that you want to get done.

‍

David Nicholson: So Rory, as we move forward with generative AI and other AI tools, what does the world of work look like for knowledge workers and for developers in particular?

‍

Rory Richardson: So I think we saw two things that were really positive. One, we were able to see that people were able to get up to speed faster. So for example, when we rolled this out internally, what we were able to see that our most junior developers, the noobs, were to learn our inferences and libraries faster. They were able to perform unit tests faster, get their documentation in, and even detect security vulnerabilities and remediation. So effectively what happened was that people were able to be more productive faster. Now, if you've been doing something a certain way all of your life, you're probably pretty fast at it. But getting those new folks on board, it can be tough. The other thing that we learned was to not necessarily focus on the hardest, most humanistic aspects of our work, but to think about all the stuff that we don't want to do anyway. Like documentation is a great example. No developer ever wants to write documentation. So if you concentrate your application of generative AI on the mucky stuff, then what we see is transformative. People are more excited about their work, they're more original, they're more creative because they're not searching through documentation that you don't really want to be doing anyway. So those two aspects have been transformative to watch. Getting developers mid-level or accelerated really quickly, and then giving them their time back to focus on the work that is most important to them and most interesting.

‍

David Nicholson: I don't come from a coding background, and so I love the idea of taking my thoughts and having code generated based on my thoughts, but I also really like the idea of then being able to look at that code and understand what I'm looking at. And so that's a very real thing that is not going to go away.

‍

Rory Richardson: Yeah. I'm a database person as well, and I would say when we started using managed databases, like in my world it's RDS, everything that's a relational database service. I was not popular with DBAs because 70% of what a DBA a does is super-duper boring, and this is coming from a place of love as a former DBA. But backups, patches, failovers are not that differentiated, not really. They're not unique to any given business. And having that taken over by managed service meant that we were shifting what a DBA actually does. I mean, these people went to school for information architecture and pushing the boundaries of insights and information, and they were doing backups. So what happened when we said, all right, well, 70% of where you're spending your time can be done by this tool, then we saw the rise of data scientists and we saw the costs of data scientists go down because there were more people available in the market that could have a perspective on information architecture that were previously unavailable or doing something that wasn't necessarily driving a lot of value back into the organization that was very unique to that person and that organization. I see the same exact things happening right now, in particular with developers, is it gives them more opportunity to innovate and to do the things that are their most human in creating something new, something different, something that solves a problem in a unique way. These are all very human things. But writing a unit test, snoozeville. I mean, seriously, no developer on Earth gets up in the morning, goes, oh, I'm going to write 20 unit tests today, yes.

‍

David Nicholson: Yeah. No one is excited about it. People cling to it if they think it's the only way they can earn a living.

‍

Rory Richardson: Right.

‍

David Nicholson: And so as long as they learn that that's not the case, we should all be okay. But what's what's on the generative AI frontier from an AWS perspective? What's coming what secret stuff? No one else is listening, just you and me, Rory. What's coming?

‍

Rory Richardson: Yeah, I'm totally not going to get fired.

‍

David Nicholson: Secret NDA.

‍

Rory Richardson: Let's take something you said earlier like is code going away because it's something that's very personal to me because I have a 15-year-old and I've been trying to teach this kid to code since he was four. I use Code Monkeys. You get the monkey to the banana and you learn Python over the course of 250 lessons. I can't make a scrapbook, but I know how to write code so I was going with my strengths on this one. Part of me's like I can't get those 10 years back. I mean, the syntax of writing Python, a specific language, when I look at my kid by the time he hits the job market, is that going to be necessary? Because, really, Python is an abstraction layer from the intent of what you want to create to ones and zeros. And what we have seen repeatedly with generative AI in production is it compresses the abstraction layers, compresses the space between intent to what you're trying to do. 

‍

So the syntax of the abstraction layers becomes less significant. So in the fullness of time, do I think the role of Python has fundamentally changed? Absolutely. The role of abstraction layers has fundamentally changed. I mean, just think about natural language. My kid's never going to learn what a gerund is, he'll never understand a dangling class participle. Because basically he has a tool now that does all of that stuff for him, and it does it really reliably. And really well just create the abstraction layer of communication. It's like, oh, here's a really nerdy metaphor, do you ever take differential equations?

‍

David Nicholson: Yes, I did.

‍

Rory Richardson: Oh. Did you fail it? Because we all failed at once. No, I'm just kidding. We didn't all fail it. I remember the day that they allowed me to use a T-65 on the test, that was four pages, and by the way-

‍

David Nicholson: First you have to learn how to use that tool. Yeah.

‍

Rory Richardson: Yeah, fair. But you also had to understand the fundamentals of differential equations, the humanistic aspect of applying a methodology or a mechanism and doing it well. Now, none of that went away. Differential equations, still around. But the four pages of handwritten mistakes that I would make on the exam, gone. That's kind of what writing with generative AI is like. It just eliminates the possibilities for mistakes with stuff that has repeatable patterns. So when I think about the future, and I am thinking pretty far out because I like to build for my kids. I want to create the world that I want my kids to live in. I love how their very mindset with technology has fundamentally changed compared to me. When I have a problem, I'm thinking, oh, I should build an app for that, right? I'm trying to make an appointment with my dentist. Wouldn't it be great if I could just go to the portal and make the appointment? My kids don't think like that. They think, oh, I would just contact one of these super agents. They would stand up an MCP server that would communicate with the agent at the dentist, and it would dynamically and ephemerally just get done. They do not think in applications, which opens the door for, well, rampant hyper-personalization. 

‍

If you are able to have an ephemeral or unique experience that is superior to a static experience, why wouldn't you do that? So I grew up with websites. You build a website and a bunch of people will come to it, and the experience isn't that different from one person to another. I mean, we've pushed the boundaries of it with companies like Amazon that have very targeted experiences. They're personalized, but we're talking about hyper-personalization where it already knows that I'm going on vacation. It already knows where I'm going on vacation. It already knows that it's a hot place and it has curated a capsule wardrobe in linen so that I'm not too hot and those are the suggestions from my concierge type experience. This trend towards hyper-personalization is going to affect not only your retail experience, but literally everything that we do moving forward, it's going to be far less static and standardized and consistent and far more idiosyncratic.

‍

David Nicholson: I love it. So what did we learn here? We learned that if you're Rory's child, not only did you have to eat your vegetables, but you also had to write code. This is amazing. Hopefully this will all turn out well with your optimistic view of the future. And I agree. I completely agree with you. I think that's where we're headed. Rory, thanks for joining us for this Enterprise AI Spotlight. Please stay connected with us on social and explore more conversations at sixfivemedia.com/summit. On behalf of Six Five Media, I'm Dave Nicholson, stay tuned for more great coverage.

‍

Disclaimer: The Six Five Summit is for information and entertainment purposes only. Over the course of this webcast, we may talk about companies that are publicly traded and we may even reference that fact and their equity share price, but please do not take anything that we say as a recommendation about what you should do with your investment dollars. We are not investment advisors, and we ask that you do not treat us as such.

‍