Transcript
If this, then that. Agentic AI is different because after Chad GPT launched with generative AI, we basically built an integrated agent-enforced platform that reasons and adapts and handles that ambiguity. And it can change course mid-tasks. Welcome to Control Alt AI. I'm Dimitri Sirota on the show. I sit down with the leading voices of what's next in AI, data, and risk. We go beyond the buzz to unpack the real-world strategies shaping the future. Your shortcut to clarity in a world built on data and driven by intelligence. Hello, everybody, and welcome back to Control Alt AI. I am thrilled to have with us one of the pioneers in Agentic, a hot, hot topic, Vivian Way, who is COO for technology for AgentForce, all things Agentic at Salesforce. So Vivian, thank you for coming on the podcast. Thank you so much for having me. Exciting to be here. Yeah. So look, lots to ask you. I think Salesforce was very, very early in embracing Agentic. And obviously you have a big, big hand in a lot of the technology and mechanics around it. So maybe before we get started, I think it'd be fantastic for the audience to just hear a little bit from you about kind of your role, what you're doing, and maybe even a little bit about AgentForce. Yeah. So I'm the COO for Unified AgentForce Platform. And what it encompasses is a combination of our trust layer and our data foundation layer, as well as the Data360, which is our product. And on top of that, we built AgentForce. So really, it's a integrated Agentic enterprise architecture all in one foundation. And what's very different is unlike a startup infrastructure and the product. Because of the integrated nature of it, it basically allows our customers to build agents and it'll act for a reason, and just like humans do, autonomously across the multiple steps through the entire ecosystem. And if you think about Copilot, it's kind of a bolt-on. It's in a different window, not exactly in your workflow. Whereas with AgentForce, it's already in the flow of work, where a lot of our customers who use sales cloud, service cloud, marketing cloud can just experience the power of agent. And maybe one natural question. So look, agents are everywhere. I'm about to attend the RSA Security Conference, this kind of annual pilgrimage of all companies in security. And my sense of it, just watching some of the previews of what people plan to launch, every single security company is now an agentic company, right? Maybe a year ago or two years ago, they were a Copilot company, but now they're all agentic. That's the topic for sure. Clearly, everyone means something a little bit different. What is your recommendation in terms of how we think about agents in terms of these kind of autonomous AIs? How autonomous are they? How integrated into kind of our daily life do they need to be? Maybe give us a little bit of a sense, just from the experience you've had with your customers, what's the right way to think about it? Yeah. If you think about earlier days, like with earlier automation, it's very rule-based, right? If this, then that. Agentic AI is different because after Chad GPT launched with generative AI, we basically built an integrated AgentForce platform that reasons and adapts and handles that ambiguity. And it can change course mid-tasks. So I think of it as like the difference between a GPS that gives direction and then one that books the flights, reroutes around the storm and reschedules your meeting if something urgent comes up. So it's responsive and it will react to the situations you're faced with. Now, some of the challenges of that is like, hey, how does the agent know the context for the user? And that's where, because of the integrated stack we built with Data360 Foundation, we basically have a single view of our customers' customers. And because of that, the agent is able to access those insights for the customers and are able to respond accordingly. So that takes us to a couple of topics here in terms of context engineering. I also just kind of came back from a Gartner data conference. And I think every company that had a catalog is now basically providing context to agentics. So maybe talk a little bit about this kind of notion of context engineering, which seems to be running side-by-side with the agentics. Yeah. If you think about it, a lot of our enterprise customers we talk to, they already have a lake house, data warehouse, data is everywhere, but they have it in one place. But what's missing is the activation. Are people using the data and activate to create value? And that's essentially what Data360 does. It's the layer on top that we have zero copy into the snowflake and data bricks of the world that allows the customers, based on the customer insights they have with our identity resolution, they're able to activate the value within the apps already. You know, some of the challenges people are faced with is if I think about the security governance, right? Governance is not a compliance checkbox anymore. It's what makes agents trustworthy enough to actually use at scale. And we built that at the very beginning of Salesforce with very strong governance foundation. So that comes, and even within the Data360, we have security and governance controls, which leverages AI to automatically look at, okay, what can users access across the different organization, depending on the role and the specific roles within our Salesforce system of record. Obviously, Salesforce has been doing kind of security and governance for a long, long time in terms of people being able to access the data store within the kind of CRM and marketing cloud and commerce cloud and so forth. What was different in terms of what you've already built in terms of kind of human kind of CRM, marketing cloud, et cetera, interaction, and now non-human, non-human identities, right? Agents coming in. So was it just the same and you just have to kind of identify that it's a non-human or were there novel things you have to introduce from a governance standpoint? Yeah. If you think about it, like now we basically created a new class of jobs that are agent managers and those agents that we have in our environment are an extension of the teams, right? So previously an SDR team of 10 people now all of a sudden has the power and the effectiveness of 100 people. And what those agents have access to is a delineation from what those agent managers have access to. Okay. And then so what do you have to do in terms of authentication, in terms of access rules, in terms of being able to define policies as to what an agent could do? And there was something else you kind of hinted at that you have kind of agents that are managers and agents that are kind of workers. I don't know if that's a formal hierarchy or that's just more of a casual observation, but I'd love to kind of hear about what you guys had to engineer to kind of support this kind of future where you have agents as substitutes or augmenters, labor augmenters to humans. Yeah. I can give you an example. Like even within our own implementation, we have like a homework that has a single view of the different data across the customer data, across the entire enterprise, whether you're in sales, service, marketing, commerce. However, if you're a leader of sitting within marketing organization, you only have access to specific information that marketing has access to that HR may not have. Right. So those are the specific governance and governance controls we have bounded by the specific orgs that you're in. And then on top of that, I was alluding to the information access, like it's not just a specific role people have access and information to, but also like even within the database itself, like certain roles, you can get into the very granularity of what they have access to, what the agents can access and cannot access. That's different from even a human that you can set limits to. So it's really policy enforcement at scale. And what ties it all together is our metadata layer. And how do you think a little bit about agents interacting with one another? A lot of what you've kind of articulated is agents going into the data, wherever the data is stored in your data lake or some proxy to other data lakes. But agents are going to be passing information, right? Or are they? Are agents actually talking to one another or they're all just interacting with the data lake independently? Yeah. We're actually moving from like task doers of the agents to the orchestration. That's on our roadmap and that's going to come. Because at some point you're going to have super agents, right? Really, from a customer standpoint, what's really powerful is you don't care whether you're talking to a support person or a salesperson. You just want your outcome to be done. Underneath it, you can have different agents. But at the top, from a customer experience perspective, it's that super agents that orchestrates across the enterprise. And that's something that we're building very actively. And where do you get, like obviously we're still early innings of kind of agentic. There's a lot of activity in this space, right? There's many frameworks, obviously Salesforce have one, but some of the other big software platforms have some. There's new tools and vendors, and obviously the big foundation models have their own kind of co-work and so forth. So what's the future? Is there going to be agents from Salesforce and agents from ServiceNow and from Anthropic and from... And how are they all going to kind of mix and interact with the data? Yeah, I think there's going to be different agents across the entire enterprise and whether it's an SAP agent or a different agent. But what we have built is essentially an open extensible platform where we could have agents and work with our agents working with other agents. In fact, earlier this year, we launched the MuleSoft agent fabric that gives CIOs a single pane of glass into all of the agents in the different environment, whether they're on an existing hyperscalers. Because that essentially is like the Switzerland of AI agents. Again, from a customer standpoint, you want the different players to be able to work effectively together. There's front end office agents that we're very good at because we already have the existing workflows from the history of the last 27 years. And there's also back office agents that's built by different vendors. But at the end of the day, from a customer standpoint, we always approach it from the customer standpoint. How do we give them a single pane of glass? And then that way they can start to govern and monitor those agents and making sure they're effective. The monitoring and the observability is going to come from right directly inside of the Salesforce platform or it's going to come from other kind of third party observability tools? We have an integrated observability stack. This was actually very important if you think about people are building agents, but unlike digital age, digital transformation, you build a website, you can kind of set it and leave it right. Those agents, they're working 24-7. And unlike humans, you can check in with your direct report once a week. But those agents, especially if they're acting on your behalf, you have to monitor them 24-7 and looking at whether they're performing to expectations. So we built observability, the eval stack, right inside the agent force, agent force studio. And that way, whether like enable our customers to tweak how the prompts are working, how the agents are acting, and then use reinforcement learning to improve the behaviors of the agent all within one place. That's amazing. Another thing I'm curious about. So one thing that's been getting a lot of attention lately is this, I don't know what you would describe it as agent 2.0, but this notion of these kind of open claw like agents, right? I think at the recent NVIDIA conference, a lot was kind of, it got calls out. And these agents, some of their skills are adaptive and kind of like self-generating, and as opposed to having to kind of pre-engineer capabilities and then just assign them tasks. How do you see that fitting in? Is that the future of agents? Or do you see that just another kind of agents that are going to be operating in a customer's environment? I've been obsessed with our Slack bot. It's not called agent yet, but it's really like, it's my assistant within my flow of work, right? So if you think about it, now I have a very personal friend who sees everything I do every day within, you know, because it integrates into your Google calendars, Google drive, et cetera. And all of it can search the information and the interactions I've had and come up with recommendations. So if you think about the evolution of that is potentially in the future, Slack could also act on my behalf if it knows me well enough and I can set certain rules, but that's kind of where I think about, you know, meeting the agents where it is, and it's a hyper personalized experience. So that's where I think the future is like in terms of the ultimate individual productivity, right? But then where the corporates are getting the most value is around institutional productivity. So that's like a genetic transformation at scale, how you completely re-imagine how, for example, sales SDR is done and service agent, how that's done, right? And we have some really good examples like, you know, with our SDR agents, I know our team, for example, in SDR used to be able to speak to like 12 to 15 prospects daily. But then after we put in the SDR agent, the meeting bookings essentially went from like 150 meetings in 30 days to 350 meetings. The agent is essentially like interacting and prospecting with customers when human beings are sleeping, right? And then when they wake up the next day, the meeting's already booked. So if you think about it, like this kind of is like a dream state where you're actually spending time interacting with other humans and the customers instead of doing all this minutiae of logistics. So what's the playbook for companies, right? So they're getting inundated with agentic messaging from everybody, security, application vendors like yourselves. Everyone has a framework. Everyone is going to support external frameworks. Agents are promising, hey, we're going to be able to reason, able to learn. Where do they begin, right? It could seem overwhelming. Where does a company get started? Yeah, I think about you start with the outcome, you know, after you understand the technology, but start with the outcome. What is the business outcome? Because ultimately the C-suites are pressured to provide either efficiency or revenue growth gains. So start with the outcome and think through across the C-360 or C-suite office, like where is the biggest bottlenecks that people are faced with day-to-day and map that workflow end-to-end before you kind of automate it. Basically, I think of agents amplify any broken processes. They don't fix that. So once you have the outcome, really understand like what are the existing processes that are broken and what can be improved and build that feedback loop from the day one so that agents can improve with usage, but only if like somebody is actually reviewing and refining. When we started about over 18 months ago, you know, we also had a different use cases. Give you an example of a sales organization. They had over 300 ideas and 300 different things, you know, agents that we could build. But what we realized is, hey, you know, from a salesperson perspective, you can only consume so much, learn so much about how to use an agent. So then we really ruthlessly prioritize, okay, these are the top two we're going to go after. And SDR is a perfect example where our website just had so much traffic that we didn't have enough human capacity to follow up with for people who come and download a webinar. And putting the SDR agent, we're able to create over $60 million of annualized type just within the matter of a few weeks after building that foundation, right? So start with the outcome. And then I think about, and truly, if you want to become an agentic enterprise across the 360, because whether it's service or sales, all that data need to be able to flow with each other, right? Again, customer intent can blow. Then you got to find somebody with like a trusted agentic enterprise architecture that's ideally integrated and open. People don't want to be locked in. And what are you finding in companies? You know, you have the who's who, including my company as clients. Are you finding that it's done decentrally where every department, let's say the SDR team, the sales team, the marketing team, the, you know, technology integration team, whatever the commerce team, they'll have their own kind of agentic or AI initiatives in general. Or are you finding companies kind of create kind of centralized committees to assess value, to assess payback on investment? Are they creating chief agents or chief AI officers to kind of assess this? What are you seeing in the field? You really see a varying degree of that kind of execution layer, right? But some of the most effective ones, they have a great operating model where IT is the center, but the CIO and their organization partners very closely with the sales organization, service organization, marketing organization. Because what's changed is, I'll give you the example. If you think about cloud transformation, it used to be like, hey, the CIO can actually just do that independently with the team. You don't necessarily need to involve so many business teams. But in the new era, you do need that trust and governance foundation. And IT is absolutely responsible for that. But the role of IT has shifted to kind of like in partnering with the different C-suite. And what's really important is, again, that data 360 foundation, where you have a single view of the customer, whether they're in the service journey, their pre-purchase journey, marketing journey. Most of all, if I follow them along the way and meet them where they are. And so that's where I find like chief data officer, CIOs, like having that layer is very important. And then on top of that, because of those agents, who do you hold accountable to the ultimate performance? It's the CROs and CSOs of the world. They were accountable for the performance. Those teams have to work cross-functionally, very closely to each other. And we have our teams that like they have daily stand-ups for our sales agents versus service agents. It's two different teams, but like, you know, it's partnering closely with CIOs. And what are you seeing people in terms of assessing the return on investment? Are they focusing on cost savings by, you know, providing alternatives to, you know, having to hire? Is it just efficiency being able to do kind of more in a smaller amount of time? Is it focused on being able to do novel things, more blue ocean, green field, you know, things you couldn't do before? How are people assessing it? Or is it a, you know, is it all three? Yeah, it's actually a combination. But what we find is the easiest to start is if you already have a bot, for example, on your website, you have the team, the capabilities to evolve that, build it into an agent. It's a much less like lift, right? And we ourselves went through that. And so we first put in the support agent. That's kind of like a no brainer. You can get it done. So a lot of companies also start with, so we generally guide our customers, depending on their use case. We usually start with like an internal facing agent that, which then quickly evolve into an external facing agent for support use cases, Q&A. And then you can evolve that into, hey, the agent can actually like help look into like a claim information and is able to respond with specific information to like resolve the ticket, right? That's kind of the agent maturity execution. But it's a combination of like efficiency. And also more exciting is like the growth opportunities, like the SDR example I gave you. Are you running into any resistance from people, right? Forget about executives. I get, you know, executives will find an RY. We talked a little bit about it. They want to find benefit. But staff, are they embracing it? Are they resistant to it? Or is it something in the middle? Yeah, it really varies. And it all depends on how leaders lead. If you build with the people and give them the context and they're transparent to people come along the way. I remember this one case where we have the agent. The agent is working well in the pilot. And but before it goes into production, the service leader said, hey, if our customers said, hey, you know, we really have to figure out the incentive system. If you think about it, a service leader used to have, you know, a few thousand people under their agreement and with agents. It's you're basically managing human and agents. In some cases, you need to figure out what are you going to do with some of the people that are have historically been doing some of the lower value work? And how do you train them to upskill them into like a different, you know, maybe a revenue generation opportunity? Right. So that's where, you know, the leader I remember was like, hey, escalate it to the CEO before we scaled it to like a 24-7 agents to a much, much bigger organization to figure out the incentive system. Hey, maybe just to finish up, we're obviously right at the throes of what feels like a revolution, right? In terms of this kind of shift from kind of purely human labor to kind of this autonomous AI powered labor. Take me a year from now, what are we going to be talking about? Is it going to be the orchestrated agents? Is it going to be something else? What's your take? A year from now? Wow. It feels like a long time from here because a lot can change. But yeah, look, a year from now, it is a long time in AI timekeeping. So what do you think is going to happen? And obviously there's no penalty if you're incorrect. I'm super excited about the idea of having a super agent. And just from a personal use perspective, if I don't have to worry about calling us, not having enough time during the day and calling a customer in the evening using voice, right? And then agent can help me resolve my issue. That is a big save. But unfortunately, not a lot of companies have put that in place, right? We're kind of in the tech world. They're talking about the evolution. But ultimately, I think the bigger impact we can create is like, hey, how do we bring more corporations along that journey and get them to actually put agents in place and let people truly benefit from that, from the power of agents? Okay, more proof points. Okay, Vivian, look, this is fantastic. Really appreciate you sharing kind of your experiences and wisdom around agentic from one of the pioneers at Salesforce and AgentForce. So thank you again. And to the audience, I'd like to say thank you for listening in. I encourage you to download and like and subscribe to Control Alt AI. And again, Vivian, thank you very much. Happy to be here. Thank you. Well, that's it for today's episode of Control Alt AI. If you liked today's conversation, make sure to subscribe so you don't miss the next one. And for more insights on AI, data, and risk, visit bigID.ai. See you next time. Transcribed by https://otter.ai
