Adam Evans at the ElevenLabs Summit

You've been extremely busy since your Palunteer days, Adam. Uh you've sold two companies to Salesforce. You now lead all of AI and Agent Force at Salesforce and you, as Victoria mentioned, are are now a part of the Time Magazine's 100 most influential people in AI. Please help us all understand how did you get here leading AI at Salesforce. Tell us a little bit about your journey. All right. Well, how much time do we have? No, just kidding. Um well, first off, thanks Ben for having me and Maddie. Incredible innovation in that keynote. Um, as far as uh, you know, my my my story or what I really love to do is I I'm I'm a builder. I like I think probably many of you in the audience like have an entrepreneurial edge or thinking about building companies and more. Uh, and right now at Salesforce, my like my MO is basically doing the same thing, but the only difference is is that it's a 75,000 person startup and we are moving as fast as we can. uh our customers uh this is like an incredible moment we're all going through as we're seeing not just with voice but bringing in AI to every aspect of the business whether it's kind of a the you know the front office or doing operations in the back end at this moment our customers are trying to go as fast as they you know ever have and we have to meet them at that moment and one of the things that um is I think really if I think about what I do on a daily basis or our challenges are what maybe Salesforce is uniquely positioned to do is we have some of the world's largest customers, these enterprise customers that although they want to go very fast, they're also operating an environment where they're used to multiple, you know, nines on everything. Their brand is on the line, the way they treat their customers and more. And although they want to be moving very fast and reinvent themselves uh in the world of AI, we have to do it in a way that they can actually uh maintain their brand, bring in the people, have agents follow policies, testing. All right, follow policies and more. I guess this is probably a little popping here. Um and so this this is where we're at. Trusted AI. And part of this is bringing in not just agents and agents that speak thanks to 11 labs uh with agent force voice uh but also follow policies are testable and uh largecale operational teams can take advantage of them. I love it and I was at Dreamforce this year there was a lot of buzz and action really around agent force. To me it really seems like this is the priority of the company and in leading Agent Force I'd just love to ask like how's it going? Uh it's going great. I mean, we've got um customers, I mean, brands like we were just watching kind of an e-commerce brand. William Sonoma is a great customer of ours and one of the things as I was watching Maddie do that demo is they have over 10,000 SKs. They're nine different brands. This is like Pottery Barn and kind of iconic things and they have multiple order management system like these are very kind of complicated at scale businesses to be able to do recommendations of products ultimately have all kinds of things from furniture to plates things just one customer fantastic customer for us. These are the kinds of customers that we have and not only have things like retail uh but another another great customer actually let's move into um uh into Europe is uh Adeco Adco if you're not familiar is one of the largest like job recruiting agencies in the world they do about 100 million applications like job seekers a year and that is a really interesting space for AI because not only is everyone's afraid of things changing from jobs but we're actually helping connect job seekers and employers employers together and in the EU that's also something where there's a lot of regulation coming out specifically around this topic. So it's one of the largest companies in the world in that space operating in an environment where there's changing regulation in a hot button issue and they're actually one of the first agent force voice um customers to roll that out as well. So um lots of things going very very well and and you kind of mentioned also um agent forces a few things for us. Agent Force is u in my part uh it is a a platform to build agents and much more. However, it's also all the technology that we're using uh internally like one of my biggest customers is Salesforce itself. So our all of our clouds, salescloud, service cloud, etc. are all going through this transformation journey as well. So they're rebuilding these massive business lines and then ultimately we're also building turnkey agents on top of that for all of our customers. So just at some level we kind of throw the agent force umbrella as almost a marketing tool around all of that. It's a platform. It's an internal transformation engine uh for us operationally and our product development and ultimately also for our customers. Amazing and and incredible customer stories as well. Uh just zooming out a bit though if you think about the just the future of the human computer interaction. How do you think about that? Well, I think that we're all, if you're in this room, you are probably are going to agree with this and you're on the forefront of it, which is everything is going to change. Uh, the way that we interact with computers is going to be conversational. I mean, it already is starting to go that way, but voice, I mean, look at the multiodality in the demos where you can now see something, interact while you're talking. It's also just human. We we've been speaking for basically since this is since our preferred modality of communication. I think the real question isn't if but it's going to be kind of the when and the how um that that transition is going to play out. Uh when I think about a lot of our like going back to kind of our customers uh so let's just talk about voice uh having uh voice on your website and a digital channel right over WebRTC kind of the interaction that you can have it opens up a lot is incredible. having it in a companion app. Incredible. However, most of our customers right now with the way they're set up and the scale of their operations, what they're doing when we go talk to the people that actually talk to customers with voice, it's in the contact center. So, what is that? That's 800 number. So, what we're doing the very first roll out isn't saying, hey, what can we do to really change kind of the uh human computer interaction? It's possible technically, but where we go is where there's interest, where there's value, the ROI story, like the foot in the water, the toe in the water to get started is the contact center, the 800 number, and the lines of business that have uh, you know, budget and goals and targets to reduce average handling time and all these other things. And that is now the first wave. However, I look at uh ultimately I think that's kind of in in my mind as a technologist I think that's great and I understand why that's it, but I really want to just do something bigger and just I really want to just kind of kill the 800 number as opposed to support it. Uh and I believe that'll be the future. However, uh even though it's technically possible, we have to bring people along and that isn't just the businesses and where they're at, that's their customers and their behavior and more as well. So I think as we kind of move into the future rapidly conversation is conversational inter interfaces to apps and everything we do is clearly coming the way and I think it's up to us as kind of product developers or if you're a decision maker bringing it your business the balance the idea of kind of the transformative more moonshot more moonshot brand new way of operating versus the kind of uh replace and incrementalism for at scale um operations you have you have to do I think both especially if you really want to ultimately compete uh better than anyone else in your market. I love it and I think we could all agree enterprise is really well known for selling um you know Salesforce is really well known for selling into the enterprise and so as you think about voice agents I'm I'm curious what are some of the biggest you know friction points and challenges that you see and taking enterprise pilots to really scale at production. Okay, so uh if you're a practitioner you know this is true. It's never been easier to build a demo It's very hard with stochastic underlying LLMs and everything like that to build something that's at scale in a production way with the 39s, right? Uh and so the hardest part of this is that the consistency and kind of ability to build agents that ultimately do a job in a way that someone let's say in that line of business like that contact center I just mentioned, they're if it doesn't work, they're not going to want to, you know, move forward with it. Their job is on the line. They're literally measured by seconds in terms of contact center. I mean, it's a very welloiled machine and understood when you get into that space. So, when you deploy agents, if they don't work or really customers don't like them or prefer the method, then it's all not going to really work out too well. It's going to take a lot longer to deploy this stuff. So, we spend I think actually I think about a lot of what we're investing in, how we spend time with our customers. I mean, uh you know, again, we have hundreds of thousands of customers. So, how do we get a way for people to take their the integrations they need to their back office systems? Of course, when it's in data cloud and sales cloud, it's much easier. However, a lot of our companies, our customers are integrated all over the place. How do we make it easier to integrate to their data to clean up their data because it's not always in the right place or they've kind of not been maintaining it over time. I mean, this the dirty messy reality. And then the business policies and the logic um like as an example that don't order more than 500 items for inventory checks. Um there's a lot of that. In fact, if I talk about William Sonoma for a second, um if you get into furniture deliveries versus plates, it's totally different policies about it. Go for delivery and all kinds of things. So there's this reality of all kinds of complexity that is in a mixture of in code uh in apps and APIs and business logic. It's written down in documents. There's tribal knowledge with Salesforce on Slack. Slack probably a lot of people are Slack users. Lots of information about actually how to conduct business is all over the place. And so part of our job or really our challenge that we have to come up with is over this next year or decade, how do we in mass take all of that information about how a business wants to run, its policies, its data, and clean that up so that we can literally just tee it up to meet the AI moment because there's this incredible technology what's possible now, but it actually has to be connected back to how businesses want to run and ultimately customers have to like it. It's got to be fast. It's got to have right tonality. all those kinds of things too with voice. Um, and that is going to be what moves the needle. I want to double click on one thing you said about data cloud because here in the Bay I feel like I've been hearing from years uh you know Mark Beni off and Salesforce in general talking about the importance of data cloud and the trust layer. Uh how does that help you really accelerate with agent force? Yeah, so uh data cloud is um a lot of things. Uh one it's a vector database. It's unstructured information. There's a lot of AI around how to index and kind of pull that information in. but also uh what it was initially built for is we call a CDP. This is a customer data platform. What this means is hey I've got a lot of data all over the place about my customers. Okay? So maybe that's I don't know data bricks, maybe that's snowflake, maybe that's kind of all over different marketing systems, what have you. And you need to kind of pull together to create a profile. And by the way, every one of our customers at scale, I mean if you're an enterprise, you're your systems aren't really clean. you've launched new product lines, acquired companies yourself. I mean, I can't tell you how many customers that we have that have, you know, 10 plus instances of Salesforce alone that they operate on. So, it's a complex environment of data all over the place. And data cloud makes it easier than ever through what we call our zero copy alliance to connect data from all these places and pull together. And then uh there's something called data graph. This is effectively a like a um both literal and fuzzy join where you're going to create a graph of all that stuff to create a profile of a customer and then it can run in real time. So as events are happening like you're clicking around on a website and looking at things uh or maybe you're having a conversation with a customer service rep and there's transcription that's happening how do we update that profile so that we can ultimately have kind of a pushbased context that goes into things like agents for better context engineering and personalization. And so data cloud uh is a massive lift for us to be able to kind of hit all of that data in one place. Kind of the harmonization of it with the graphs, the real-time kind of push for context uh and of course the unstructured data with kind of the vector database and innovation on that makes a lot of sense. I I imagine as you're talking with agent force customers and prospects, there's probably this wide range of people from AI tourists to like ready to deploy things at scale. For those that have really leaned in and move faster, what what do you think separates them? what's their approach and and what's different about those organizations? Yeah, who who goes the fastest? Um, okay. So, u I think probably two things. One, I kind of covered a little bit in terms of have you been paying like taking your vitamins? Have you been cleaning your data up? Is your knowledge in good shape? Do you have APIs for all your systems? Have you been consolidating? This is not a an attractive thing. This is just the reality. By the way, if you know AI and agents are the new apps, great. Yes. However, software integrating to all those systems, you still have some of the same problems that we had before just because the new technology is amazing. Well, how good is your source data, right? And when we, by the way, we turned on uh one of the first agents we launched, we call it when we use our own uh products, we we treat ourselves as customer zero. Um we have uh help.salforce.com. We turn on our customer service to help people, you know, change my password. I need to know how to set up a user. Salesforce has a lot of products. It's like 500,000 articles by the way on support articles and um the the operations by the way fantastic. We've saved actually 500,000 uh hours on this. We you know it's it's got an 80% resolution rate. It's really great story. However, when we turned that on we were like why is the agent saying that it's a bug lm hallucination? And the answer is no no actually that's what the data has always said in the article is just it was like we turned a flashlight on and now we could see some of the problems. And I'm saying that because this is probably the biggest hurdle for a lot of organizations that they just have to invest more in this. And by the way, when we launch agents in these things, there's a good reason to now before they could kind of get by. And just like us internally, we saw it. There's that kind of debt payown kind of investment people need to make. Um, and the second thing which is probably uh maybe that's more of a a kind of a weight for you. The other one is more of kind of the top down kind of push or maybe the carrot which is the fastest moving customers that we have um have um come with a mandate from the CEO a top down like we are going to transform ourselves in fact that a deco customer I just mentioned um they're leaning into this moment to completely rebuild in fact we have a joint venture that we've announced with them they're really going into how do we think about employment in the future with AI uh and when you have um more visionary leaders that are willing to kind of go all in and especially if they're large companies, you can you could spend years not moving forward and having all kinds of reason to deliberate and things like this. And it kind of comes down to somebody coming in and saying we're going to ship this in a month or a quarter and this is just what we're going to do. And so I see both I would say the the maintenance and the readiness on one side with the sense of especially for large companies the thought leadership and kind of galvanizing support. Those two things I think are what I've seen is probably the most uh the combination of what really unlocks um many of our customers. How do you think about the the current architecture of you know voice in in terms of like the cascading model and orchestration versus a voicetovoice model and and what you've seen in the space as I'm I'm sure you've evaluated it. So I'm really excited for voice to voice. It's fast. It's natural. You really great birectional. It's like in in some clear the future. However, what I was just describing is the challenges to get to the multiple nines of consistency, that is really where the rubber hits the road right now. So, the cascading approach and for those that don't understand what that term means, um I just define it as basically kind of the ears, the brain, and the mouth. Like you've got your ASR, you got your LLM, and you got your TTS, right? So, if that makes sense. uh and the division of these things although can introduce more latency and some other things like that and has some uh drawbacks um I think right now is the is where things are at state-of-the-art you need control in the middle in fact as we have moved to do more and more releases uh with working with our customers uh having more you know in some kind of weird twist of irony more determinism inside of the agents looping and reasoning and more control over certainty um is what our customers want versus just hey the model can do everything just toss it over like prompts alone are basically not enough. If you come into uh regulated environments with we work with a lot of banks and insurance companies and more they need certainty around policies right and so um you need more control. So although you know speech to speech or voice to voice um looks amazing right now instruction following and these things I think are not um not where they need to be uh in order to really take this to the mainstream at an enterprise level. Maybe if you're creating an interaction in uh more of a social thing or a game or more just conversational or interesting that's your product fantastic. uh for most of our customers that want to do transactions at scale and delight their customers and turn value um they need something a little bit more concrete and that's where the cascade approach uh for us uh absolutely makes sense. Really really appreciate that that insight. Okay. So what role has 11 labs played in your agent force journey? Well so 11 labs is a great partner. Um agent 11 labs is the voice of agent force voice. So a really pivotal role is the answer to that. Uh in terms of 11 Labs, we did um an evaluation as one would do of a lot of the different players that are out there. Uh what did we find? Basically, the combination of all things pointed to 11 Labs, quality of voice, speed, ability to have kind of right tunability in the knobs, ability to have uh you know better word boosting and more as we do a lot of things in industry. So, um, being able to pronounce certain things correctly, uh, whether you're like fixing a a wind turbine or you're deep in an elevator in field service is a multi-billion dollar business for us. There's a lot of words and things. 11 Labs perform very well uh compared to others. And then also, I think the uh just the quality of um the voices and thinking about brands. I think we're early innings right now. I love the voice uh uh market. Is that what it's called? Voice library. Sorry. um of more and more voices getting out. I think this is like a really cool almost like a day two problem where as everyone launches their voice agent and it's the frontier of their brand, they're going to want to differentiate right on how they sound uh within their space. Um so that's really fantastic having high quality and having that diversity of voices coming in. Um and of course latency and those are all I think the most important criteria for us and why we chose 11 Labs. Thanks Adam. Yeah, so humbled to play a part. So last question now as as you look at the future here I mean it feels like our pace of progress is actually accelerating and since even the the launch of agent force and and in the broader space the the innovation is just really accelerating. I'd love to hear from you what what you think is you know the future of conversational uh you know intelligence and agents over the next 5 10 years. Okay. Okay. Well, five or 10 years is a long time in AI, but I I would say look, we're we already kind of covered something that the way that we interact with uh computing probably much more voice and conversational. It's natural. The multimodal aspect, I think that a lot that's just going to get tighter and tighter in terms of the interaction space, visual and audio auditory. Um I don't know. I think that uh right now one of the other things that we're thinking about is uh you know personal AI uh versus like basically actually I think this is probably a big thing is how will people interact I think about brands a lot our customer and our customers customers how are people going to interact with brands in the future and um I think there's kind of there's two ways one is that there is a direct kind of channel to them like you go to their store, you go to their website, you call their phone number, like you're kind of typing their domain and you uh go there directly. And for that, I think voice and conversational interface is basically an agent. I think more broadly speaking is it's like the the next channel, the new channel. And what what do I mean by that? I mean that, you know, just to date myself a little bit, when the you know, the web came out, uh the internet came out and everybody was basically saying, well, do I need a website or not? And like you kind of know how that played out. It turns out your website is like just a different kind of storefront. It is a new way that you have to think about and build or and now we've got mobile like your companion app to your business. It's like critical. And I think that's one big way that companies need to be able to build agents that are their brand ambassadors and they need to be able to meet the moment for conversation and personalization and context and all the things that we know. And I also think that um we're going to end up having personal uh AI as well. And this gets a little bit more into the ATA type stuff, the agent to agent type stuff where you are going to interact with uh brands almost in a more indirect way by you know we're we're seeing this happen kind of fairly real time the last 6 to 12 months already but I think that's going to be a whole different kind of interaction uh modality as well for companies. uh maybe more competitive uh maybe less control uh but that I think is going to be part of where we're so voice clearly and I think also AI as proxy is going to be something that we're going to probably start seeing a lot more at the consumer level uh in maybe a lot less than 5 years actually um kind of roll out. So I'd throw that out there. I love it. That's a future I want to be a part of. I remember when the web came out as well. I was day trading tech stocks in high school and I've just always fundamentally believed technology is going to have, you know, a a real uh upside. There's so much potential in the world to impact it in a positive way and Agent Force is such a big part of that. Humbled to play a part of that journey as well. Ladies and gentlemen, please give it up for Adam Adam Evans. Thank you so much. Thank you guys. Yeah. Thanks so much. Thanks.