How agentic AI will save customer service, with Bret Taylor

0:00Starting a business comes with its share of ups and downs, which is why staying true to your vision is essential. a non-negotiable for Romeo and Milka Bregali, Capital One business customers and co-owners of Ra's plant-based restaurant in New York. Romeo and Milka took a leap of faith when starting their own restaurant, gutting an empty space and building it from the ground up. Every pipe, every wall, every detail. But building from scratch came with a heavy financial burden, which is when they turned to their Capital One business card. With the flexibility of the card's no preset spending limit, they were able to spend more and earn more rewards while bringing their vision to life.

0:36Today, Raz's success is proof that with passion and the right support, it's possible to make your dreams a reality. Learn more at CapitalOne.com slash business cards. Pioneers of AI is made possible with support from Inflection AI. It's not just enterprise AI. It's your enterprise AI. I think if you surveyed your listeners and say, do you like chatbots? my guess is zero out of 100 of your listeners would say yes, because what they associate with chatbots are these annoying robotic pieces of friction. You just want to talk to a real person, right? I think if you surveyed those same 100 people and said you like ChatGPT, they'd probably all say yes.

1:22Brett Taylor says that the difference between the chatbots of yesterday and ChatGPT is a qualitative one. How did talking to this chatbot feel? Was it enjoyable? even delightful? Or was it scripted, lifeless, and unhelpful? Brett is trying to make the kind of chatbots that fit squarely in the delightful, or at least actually helpful, category. He and his company, Sierra AI, are partnering with brands like Weight Watchers and Sonos to create customer service chatbots that feel human. It's hard to find a Silicon Valley track record that matches breads. He's worked with some of the biggest names in tech, helping develop ubiquitous tools like Google Maps.

2:08He was also CTO at Facebook, now Meta, co-CEO at Salesforce, and now serves as chairperson at OpenAI, the company behind ChatGPT. Today, our conversation with him is focused on the future of AI customer service agents, AI safety, and of course, his role at OpenAI. I'm Rana El-Khalyubi, and this is Pioneers of AI, a podcast taking you behind the scenes of the AI revolution.

2:49Well, welcome to the show, Brett. Thank you for being with us today. Thanks for having me. So I like to start my conversations just kind of learning about how you're using AI in your everyday life. I use AI almost every day. And my son, who's 15, he loves being at the forefront of all these AI tools. And he asks ChatGPT for recipes about dinner each night and whatnot. So how do you and your family use AI? I use ChatGPT 20 plus times a day. So kind of for everything. So I use it at work when I'm writing. I use it, a product called Cursor, when I'm coding at home. I've got three children as well.

3:27So my daughter started reading Shakespeare. She was asking me what a passage meant. I wasn't very good the first time I read Shakespeare. Put that into chat GPT. We got some great answers out of it. How old are your kids? They're sort of early teenagers. So it's been great. I mean, it's sort of a co-pilot to my life. It's like the expert I didn't know I needed in pretty much every domain. So it's almost like saying like how to use Google. It's not like anything. It's everything. I've gotten to the point where I couldn't imagine working without it. Yeah, yeah, similar. And you go to it for like just thought partner kind of stuff, right?

4:03Like, what should I think about this? The thing I'm most excited about is I always thought of computers, I think most people did, as kind of these utilities. Looking at facts, you know, processing numbers really quickly, databases, very predictable sort of tools for automation. But now computers are creative foils. My daughter was doing an art project for one of her classes and was having basically whatever the artist equivalent of writer's block is. And we used ChatGPT to come up with some ideas. And I think that's a really compelling evolution of how we use technology, that it's not just a source of facts and automation, but actually something that's a source of creativity.

4:45And I think that's pretty exciting. It's very different than how I think most people conceived of computers and software even like a couple of years ago. Yeah, absolutely. So we're here to talk about AI agents in Sierra, but before we get into that, our listeners are going to be thinking about your role as chairperson of OpenAI. So I want to kind of tackle that first. So as chairperson, you don't make the day-to-day operational decisions in the company, but you do have fiduciary responsibility. 200 million people use ChatGPT each week. So that's a big responsibility. How do you think about that?

5:19How do you think about your role? Well, OpenAI is an unusual organization compared to most in the AI space because it's a nonprofit. So what that means in practice for the board in particular, but for the company broadly is it's driven by its mission exclusively. So the mission of OpenAI is to ensure that artificial general intelligence benefits all of humanity, which is pretty heady. I mean, it's a big mission. It's very broad. That's how I think about it. So, you know, when I think about recruiting the board, which I played a part in this past year, you think about all the different areas of expertise you want represented when you think about what does it mean for AGI to benefit humanity?

6:03How does it impact the economy? We've got Larry Summers, who's one of the world's great economists. It's about cybersecurity. Paul Nakasone, who ran the National Security Administration, is on the board. It's safety. Zico Coulter is a professor at CMU and an expert in AI safety is on the board. Sue Desmond-Heldman, who is the head of the Bill and Melinda Gates Foundation, is also on our board. It's a nonprofit, so we think about what does it mean to really go beyond building AGI but distributing those benefits. So really trying to, at the board level, have all the different areas of expertise to fully represent that mission.

6:44And, you know, I don't want to pretend we're perfect, but I think it's really fun to work at a company that cares so deeply, not just about building a technology, but distributing its benefits, ensuring it's safe. And I'm really grateful to be a part of it. Yeah. Recently, Nobel Prize winner and computer scientist Jeffrey Hinton raised some concerns that open AI is focused less on safety and perhaps more on profit. How do you balance profit and purpose for open AI? I think OpenAI is about purpose. So I'm not sure for me that's a question. There's a lot of nuance, though, about what does it mean to develop artificial general intelligence responsibly?

7:28And I think there's some real substantive disagreement in the world about what that means. And I think it's important, especially as a board member, that I listen to all those viewpoints just because I think we need to have the humility to understand not every decision we make might be right. And so it's very important and I welcome feedback from experts like Jeffrey Hinton, who's a remarkable researcher. We would not be here if not for his achievements in AI, but also stakeholders in government and elsewhere. Broadly speaking, I would characterize OpenA's approach as responsible iterative deployment.

8:04The idea that there's actually lots of elements of benefiting humanity beyond safety in an ivory tower. It's unlikely we can contemplate all the first and second and third order effects of technology as nuanced and powerful as AGI. It's unlikely we'll be able to contemplate all the different ways one could jailbreak a model. It's unlikely we could predict the impact on jobs without sort of having the technology available in some form. Similarly, you know, I think if I look at the products that OpenAI makes, I think they're probably some of the most powerful ways to ensure that AGI benefits humanity.

8:42We were talking about ChatGPT. That is probably the most concrete manifestation of AI in people's lives right now. And it's probably doing more to benefit humanity than any product as it relates to AI that I can think of. So I don't want to dismiss such a renowned scientist's concerns at all and I listen deeply, but I do think that for OpenAI is about purpose and recognizing that, you know, we want to build this remarkable technology in a responsible way and have really critical and thoughtful conversations about what it means to pursue our mission. Yeah, absolutely. So last question on OpenAI.

9:21Recently, a few executives have left and I can see how that could be concerning for the company, but I actually have a different perspective. I think it's pretty awesome that this core set of founders of this organization are now potentially off doing their own flavors of AI companies. So I don't know. How do you think about that? Silicon Valley has been defined by the fluidity of talent. If you look at the role of Fairchild Semiconductor and the folks who started some of the seminal companies that gave Silicon Valley its name, If you look at the authors of the Transformers paper, most of whom went off to create other things.

10:03Fairchild Semiconductor was a seminal company creating microchips, among other things. And their work led to Silicon Valley's name. In fact, some of the company's founders went on to start household companies like Intel and AMD. The Transformers paper Brett is talking about was written by researchers at Google, and it kick-started the rise of generative AI. The PayPal mafia is another example that comes to mind. These former founders and employees of PayPal went off to start companies like Tesla, LinkedIn, YouTube, and the list goes on and on. I guess the point is, there's no monopoly on talent.

10:45I actually just start with gratitude. OpenAI would not be here if not for all those people who are currently there and left. And I think I wouldn't be in this chair if not for the creation of GPT-3, GPT-4, ChatGPT, and it's transformed society. So I'm just grateful for everyone who helped build OpenAI into what it is. And also grateful for all the new amazing folks coming into the organization as well. So this is for those folks outside of Silicon Valley, my sense is it's hard to understand how much this is just the way the culture works here. And I think it's really exciting. I think that when a technology is so meaningfully different than what preceded it, I think that you really do need a large number of entrepreneurs exploring the impact of that technology.

11:34Not everyone's theories will be right. And I'm just grateful to be here and be a part of it.

11:43We're going to take a short break. When we come back, we dig into the future of agentic AI and hear about a particularly charming AI customer service agent who goes by the name Duncan Smothers. Stay with us.

12:08If you've spent any time building AI products or leading technical teams, you know this. Transformation doesn't fail because of ideas. It fails because teams can't move together. Enter Atlassian's Teamwork Collection. It has planning in JIRA, documentation in Confluence, video updates in Loom, and now AI agents in Rovo, which connects the dots across your work so nothing gets lost. It's one AI-powered teamwork platform designed for how modern teams actually build. Learn more at Atlassian.com slash TeamChanger. That's A-T-L-A-S-S-I-A-N dot com slash TeamChanger. So now that we've covered OpenAI, let's dig into Agentec AI and Sierra AI, your new company.

13:03You've done so many things over the course of your career. You were the co-creator of Google Maps. You started and sold a company to Facebook. So you spent some time there. You started another company and sold that to Salesforce and became the co-CEO. And you were on Twitter's board and now that you're a person of OpenAI. So when you decide to start something new, it's no small thing. I would love to hear the origin story of Sierra. And what was that tipping point where you were like, you know what? That's going to be my next thing. Yeah, I'll start with just how I view the AI market and then tell you why we focused on Sierra.

13:38I think that right now we're being driven by changes in technology. And a lot of the applications of those technology probably are obvious, but they don't exist yet. And it reminds me of the early days of the web browser. I think when the web browser first came out and the internet was new, the idea that one could purchase things online was not conceptually complex. Getting to the point where that was possible, how do you actually use a credit card? How does that work? How do you actually sort of fulfill the order? What are the last mile logistics? Similarly, the idea that with all this information online, you'd want to search over it wasn't particularly novel.

14:19Google was not the first search engine I used. It just turned out to be the best and earned this market. And I think the same is true of a lot of the defining companies of the web era. So now move forward to artificial intelligence. I think there's a lot of areas that intellectually, you know, are going to be impacted. Education, the law, marketing, customer service, customer experience. So Clay Beauvoir, my co-founder and I spent a lot of time thinking about what are the areas that will be impacted first? what are areas that need scientific advances to really work and what are areas that can work with current technology and we really settled on cera and i'll tell you in a short sort of short version of what we do so we help companies build customer facing ai agents so if you have a problem with your sonos speaker you'll chat with an ai now powered by our platform to help you fix it if you have a series xm subscription and you want to upgrade or downgrade your subscription you'll chat with Harmony, their AI agent.

15:20If you buy a pair of Olakai shoes, which is one of my favorite flip-flop brands, and heaven forbid you want to exchange it for another pair, you'll chat with their AI agent. If you're a Weight Watchers member, there's a tab in their app called 24-7 Live Coaching, and it's now powered by AI and built on our platform. And so broadly speaking, we're helping companies build that sort of frontline customer experience that's now conversational. And in the short term, it's helping a lot of people improve customer service, which is often a sore point for many brands. And often it's very expensive to operate and hard to provide the quality that I think brands want to.

15:56But it's also a new form of customer experience. I think that I imagine in 2007, when Steve Jobs introduced the iPhone, few of us would have contemplated that we'd send nearly 100 % of our email by thumb typing on a touchscreen, but we do. and we do not because it's better than a keyboard. We do because it's the most convenient way to do it. It's in our pocket. Our thesis is the same thing will happen with conversational experiences, that in five or 10 years, the main way we will interact digitally is by having a conversation with AI. And I think your AI agent will become just as important as your website, just as important as your mobile app, perhaps more important because of the convenience of it.

16:37So we're really trying to help brands embrace that future, help them in the short term with things like customer service, but long term, how do they build their AI agent that will help their customers do everything? Everything. Yeah, I can imagine, you know, Harmony, you mentioned Harmony, SteersXM's AI agent. That could be your interface to the company, right? It could help you with debugging questions, but maybe like advice on what products you could buy or, right? Well, just think about the services you use, not daily, but maybe monthly or yearly. Maybe you are filing an insurance claim because you got a fender bender.

17:15You know, maybe you want to move. It's like right now it's painful to do that, right? It can be. And it's particular, you know, no matter how great you make those digital experiences, you have to maybe install the app that you didn't have installed. You have to sign in and figure out what your password is. Then you have to figure out how to navigate it. And you're in a moment of distress. you don't necessarily know how the UI designer designed those buttons and you're figuring it out for the first time. And hopefully they did a great job designing it. Imagine just to be able to say, Hey, I just got a little fender bender in a parking lot.

17:49I need to file a claim and a helpful multilingual empathetic agent can just guide you through that process. Maybe you actually don't want to chat because you're standing on the side of a highway. You can do it over the phone and you don't need to wait on hold because AI answers instantly. I think the idea in those moments that matter that you have to wait on hold, that you have to be transferred to the right person who has that expertise, it can be such an inhumane experience for people who do it. So I'm just so excited for companies being able to provide this white glove, incredibly empathetic experience 24-7 and also at a fraction of the cost of existing options.

18:30I love that you brought up empathy. I spent 20 plus years of my life building emotion recognition and empathy into machines. And I think it's really important that these AI agents have that empathy because that's going to be the thing that builds trust with these agents. Do you agree? Absolutely. And, you know, the way, you know, the earlier chat models were developed was with this technique called instruction tuning. Instruction tuning. It's a technique used to fine tune large language models to make them more accurate. It's remarkable how empathetic most conversational models are just out of the box.

19:09And they're sort of designed to reflect the sentiment of the person having a conversation with them. And I think that there's a lot of things that actually are afforded by technology that are counterintuitive. So I brought up, you know, maybe you're having trouble with your Sonos speaker. Not everyone's as geeky as you or I are, right? So like for some people, they might want to be on the phone for 30 or 40 minutes. If you're perhaps maybe in a contact center doing that, maybe your manager wouldn't be so excited about you spending 40 minutes on the phone. Maybe you need to, not in a rude way, but you need to sort of move it along.

19:46Well, then AI, take your time, you know, no big deal. And I think for all these conversations, I think that you can have not only the AI be empathetic in terms of tone and substance, but there's more subtle forms of empathy. like being able to answer the phone instantly, being able to operate at your cadence and your pace, not necessarily the pace that's oriented by sort of other business interests. And I think that's incredibly exciting. And I think there's also a more subtle thing. There's a website, I can't remember the name of it, where you can find the phone numbers for consumer brands, because it's so hard to actually call most consumer brands.

20:22I think it's based on a real simple business truth, which is it's very expensive to have a conversation with your customers. And as a consequence, most particularly consumer brands, you know, make it as sort of the last line of defense, not because they don't want to talk to you. I think every brand would love to have conversations with every individual customer. They just can't afford it. It's just not cost effective. With AI, conversational AI, if you can bring down the cost of a customer conversation by one or even two orders of magnitude, how many more conversations can you have with your customers?

20:56And that's really exciting in my mind because you can actually have the experience of, say, going into a retail store and talking to the best associate to help you, guide you to your purchase. In those moments that matter, like we talked about, you can have someone that's incredibly empathetic with instant access to information to solve your problem instantly. So I think going back to the second and third order effects, I think it will really change the way companies design their customer experience. And I think we will orient around personalized conversations. And I think that's really exciting.

21:30So brand identity is super crucial in these cases, right? You're not going to give SiriusXM the same AI agent you're going to build for Weight Watchers. So how do you approach that? How do you get the right brand voice? We believe that AI agents should be brand ambassadors. So we spend a lot of time training our AI agents, not just on the standard operating procedures of your company, but what does it mean to represent your brand? One of the ones that I find really delightful, if you know Chubby's, the Shorts brand, their AI agent is named Duncan Smothers. And he's got a little sarcasm. He's a little fun, you know, like it's really fun.

22:10I love the Chubby's brand. They've just invested a lot in their brand. And you can feel it. when you're chatting with Duncan and if you chat with Olakai's agent, the Hawaiian-inspired brand, you'll see it and feel it in those conversations. And I think that's great. And I actually think that just like you spend a lot of time when you're designing, say, a website, it's not just about the functionality of the website. What do you feel when you see it? What's the color scheme? And that is a big part of design. Your conversational AI is a brand ambassador. So whatever your brand's values are, it should emanate from those conversations.

22:43And that's really, really exciting. And it's probably one of the most fun parts of designing these AI agents for their customers. Yeah. Can we geek out for a moment? What is the sort of training data you, you know, this brand kind of brand specific training data that you use to customize these agents? Like, how would you customize these agents? So a lot of the companies we work with actually have brand books and voice guidelines already. So particularly in retail, a lot of companies do that. And if you're trying to train your associates on how to represent your brands, we work with one luxury furniture retailer and they have guidelines on sort of the language of luxury and what that means.

23:23I learned so much through that process. So fundamentally, we take all the training materials you provide, your best associates on it. And then the other thing that we do is we help companies tune it over time. So for those people who are listening right now who ever made a mobile app or a website, you have all your ideas and then it meets reality, right? And you spend a lot of time, maybe you had a great idea for one screen in your app, and then you did a usability study and you found out no one can figure out what to do. The same can happen with conversation design too. So we spend a lot of time tuning, tone, flow, and we built a lot of tools for customer experience leaders to actually iteratively improve and evolve their agent over time.

24:04One of the interesting, I would say, almost philosophical points about building a customer experience with AI versus building, say, a website or a mobile app is it's not really bounded. So just think of a typical website. Let's just say a typical retail website. You could probably click and go to every single page on the site if you had enough time. With a conversational agent, whether it's over the phone or over chat, it's just a free form text box. Your customers will say whatever is on their mind. And so I would say consumers and customers have a little bit more say in your customer experience with conversational AI, and they'll take it in directions that you didn't foresee.

24:42And so I think as you're thinking about brand, but also as you're thinking about tone and substance, like what is this AI capable of doing? One of the most fun parts of developing these agents with particularly large brands is they'll find things their customers want to talk to them about that they didn't foresee. and actually changes what they develop with their AI. And you take all that data and you basically iteratively improve or tweak the agent all the time. Yeah, and it's very organic. And I think that it's impossible to predict everything your customers will want to talk to you about. So it's not just getting it right from day one, it's actually building a feedback loop.

25:22And a lot of our customers call their customer experience leaders who work on their AI agent, the AI architects. And I think it's a new role. Conversation design is the new UI design. AI architect is like application, sort of the product manager for the AI agent. And, you know, I'm really excited for these new roles to emerge and excited to build the tools for those new roles as well that are evolving over time. And, you know, I think we have a fuzzy glimpse of what all those roles will be. And I think just like the emergence of the web where, you know, what a web designer is, what a web developer is front end, back end.