David Schmaier, Salesforce | Dreamforce 2024

It's Agent Force, Agent Force, and Agent Force. So we're going all in on humans and AI agents working together to help customers. And we really think this is a big story because a lot of companies have been trying to build their own AI, and we don't think this is something you want to do yourself. If you go to a surgeon, you don't do it yourself; you go to somebody who really knows what they're doing. And what we've done is built this out of the box so you can literally turn our agents on in hours and not take months or even years to build the AI yourself.

It's a great question, George. And we really have built Agent Force on top of the foundation we built over 25 years: multi-tenant in the cloud, proven, scalable, trusted. On top of that, we built something new in the last several years called our Data Cloud, which allows you to basically integrate, harmonize, and unify all the data together. And our view is the data powers the AI and the agents. The three of us all work together, and we put data into the system, and the system magically does things for us.

Sure. We've been working on AI for over a decade at Salesforce. We have a Stanford professor of AI and machine learning who runs our Research Center. So we've been working on it, and we decided to double down and go all in on agents and AI, and the flow work. That's what we've done. And what I'm most excited about are the customers. Customers like OpenTable, Disney, ADP, Wy, and Saks Fifth Avenue. You're going to hear from them this week. You're going to hear their stories. They're going to talk about how the agents have literally transformed and revolutionized the way they interact with their customers, so they can do what has really been our vision for 25 years: help companies connect with their customers in entirely new ways. And Agent Force is definitely an entirely new way.

That's exactly right. I think it's very, very well said. So we have always had the full Customer 360 view. And some of the other CRM players give you what they call a data lake. The Data Cloud takes it further because we can bring in SAP data, we can bring in billing data. If somebody clicks on a certain product page on your website, you instantly get that signal that we can send to the right salespeople and say, 'Hey, somebody just clicked on this page; they might want to buy this particular product.' That would be good to know instantaneously because leads are sometimes perishable. So we've sort of connected all this together. And the amazing thing about Agent Force is we have over 200,000 customers at Salesforce, and it's all built on these core fundamental technologies at the data layer and at the workflow layer. Our flow automation layer does 44 billion workflows a day, and so it's used by millions of people. We're going to turn AI agents on for people right at Dreamforce. And we had our own people turning on agents in one to two hours, working really smart, intelligent conversational agents in one to two hours.

Well, the data lake, our Data Cloud, allows you to... we now have over 400 connectors, and with MuleSoft, which is the number one API management system, we can get you to thousands and thousands of applications. So we can virtually connect to anything. But we can also connect to other data lakes like Snowflake, Databricks, Redshift, BigQuery, and Azure Fabric data. So we can connect to any of these other data lakes. And we just announced a big partnership with Workday, with IBM, and other big partners that are also joining what we call the Zero Copy Alliance, where you don't need to copy all the data into our Data Cloud; we can just point to it but give you a logical view of that.

Sure. We have something called the AppExchange. And so we take a platform-centric and ecosystem-centric view to everything we do. I'm the Chief Product Officer, and Mark and I have a lineage going way back working at Oracle before Salesforce together, so we've known each other for a long time. And I'm a big believer in this platform-centric approach. The AppExchange today has over 6,000 business applications. And what we just announced is a new agent ecosystem where now you can take our Agent Force platform and you can build your own agents. That's what Workday is doing for employee onboarding or employee service. They're building with us on the platform because they are really the system of record for employee data. We're the system of record for customer data. Now we bring both of those together on Agent Force.

I would say it this way: potentially. I think we're the new way of bringing agents to the enterprise. So the old way was many of the hyperscalers and many of the AI companies would say, 'Hey, you, George, you need to train your own models, and you need to build your own RAG, and you need to get a vector database, and you too can be a computer hobbyist and plug this all in together.' And what we're great at doing at Salesforce is making hard things simple. And we've done that for 25 years. Analytics did it with Tableau. We can bring, like, this laptop. If you're a Salesforce customer, we'll have a couple experts, it's kind of like the Apple Genius Bar. They'll turn your agent on in one to two hours, and they'll show you how to do it. So when you leave there, it'll be in your sandbox. You can go home and test it and look at all the use cases. And then later, if you want to go live with it, you can go live with it. So it's very, very cool. We believe that we've been in this computer hobbyist era of AI. That's why these projects have taken a long time. That's why the returns have not been what everybody expected. And we think this is the new chapter. We think it's going to be easy.

Yeah, let me give you a couple of tangible examples. The first is Wy, the textbook company. They're going to be in our keynote. They've been a Salesforce customer for about 10 years. They have a huge spike in demand when our kids all bought their textbooks. But we probably bought them from them two years ago. This is back to school season, so they need all these seasonal workers. But now they don't need them because they're using Agent Force. And so they can handle the demand without hiring all those seasonal workers. Another example is OpenTable. They were going through a sort of a chatbot bake-off, and they're a big Salesforce customer. They said, 'Oh wow, we didn't know about Agent Force. Can you tell us more about that?' And they quickly standardized on Agent Force because chatbots are very brittle. You had to program if-then-else, the branching of like if the customer says this, then do this, or else do that. Now with the LLM, you don't have to do any of that. It can reason and it can plan. And so we're seeing incredible results at OpenTable. And then Disney, which has 17 million visitors a year to the theme parks, they're using Agent Force. You know, Andor and Obi-Wan is the next series that you should watch on Disney Plus because it knows who you are, it knows what your preferences are, it knows everything about you, and it can do it in a very human-centric way. And if you want to talk to somebody too, you can. We support humans and agents working together. So it's not either/or; we think it's both. We think that there are some situations where you're going to want to talk to people. Like in the theme park, there might be somebody with an iPad guiding you to the next ride that's using Salesforce. But we think the intelligence and the smarts and the AI make that all effortless. It makes it the experience that you want. It's what we call AI that you would really hope to have, an AI that you imagine.

Absolutely. We think that we have certain domain expertise. Our namesake product, our Sales Cloud, is the world's number one sales force automation product. Our Service Cloud is our biggest product in the entire company; it's the number one service product. And Marketing Cloud is also number one. But people can build agents too, and if they can build a better one, great. But we have in the AppExchange, 90% plus of our customers use one or more AppExchange products because it's impossible for us to build everything. A good example is the Workday example. We're not experts at employee onboarding; Workday is. So why try to be in the game? They're a great partner. We use their products; they use ours as a customer. The same with ADP and payroll. Why not let the experts with the domain expertise build the knowledge and the skills? And if you go to our website, you'll see our skills library and what we call our Agent Force skills.

I would say it's more of an evolution than a pivot. We've always been a platform company. From the very beginning, you could customize Salesforce. You could add tabs or change things. And now it's gone way beyond that with Flow and MuleSoft and other capabilities. So you can literally customize everything if you want to. The magic part is we automatically upgrade every single customer. We do three major releases a year, and everybody gets it, and it literally just works. There's no big IT upgrade project. So we can keep our 12,000 engineers building more and more, and customers just get it and they can instantly use it because of this multi-tenant cloud model. So it's really amazing. But we've been a platform company since the very beginning. I think what the definition of platform has expanded dramatically. So before, it was about objects; then it was about flows. Now it's about agents. But it's the same platform-centric, ecosystem-centric view of the world.

The biggest companies in the world use Salesforce, and small ones and medium-sized companies too. But we'll typically meet with the CEO. And the CEOs are all saying, 'How do I use AI for competitive advantage?' If you come to San Francisco, Uber was a revelation, and they're a huge customer and a great example of how you disrupt an industry. And they run their whole company on Salesforce. Amazon's another great company, and they're one of our biggest customers in the world, and they disrupted their industry, and they run their whole company based on Salesforce. But you might see all these Waymo cars now and say there's a new way to get around through these autonomous vehicles. I took one yesterday. The CEO says, 'I want to be the Waymo of our industry, or the Uber of our industry. Let's do it.' And then he turns to his CIO or CTO and says, 'Make it happen.' And that's where we start spending a lot of our time. And there, I think, there's been a little bit of early attempts at this with DIY AI, do-it-yourself AI, where people are trying to cobble this together and build it themselves. But it's pretty hard. We have a PhD from Stanford in machine learning and AI who runs our AI research lab, and we have a team of PhDs who work for him. It's pretty hard to feel a team like that. Some companies have that, but most don't.

Yeah, we're in a brand new world. In fact, they don't run on Mistral. We have what's called an open LLM framework. So actually, you can pick the LLM of your choice, and we have certain ones that we've found work best. And Agent Force is a perfect example. We're beating the results people are seeing with the custom application of LLMs. And the reason we're beating it is because we understand all the data, we understand all the metadata, which is the real semantic meaning of what's really going on. Therefore, we understand the context of every single customer interaction across every channel. So it's really not the LLM that's the magic; it's putting it all together with Data Cloud. And we've also introduced with Agent Force something we call our Atlas planning and reasoning engine. This is really cool. Our AI research team, run by this Stanford professor, built this. And we built this to plan the outcome of what the actions are according to the action library, according to this customer profile. So you don't need to build all this stuff anymore. That was like chapter one of a 20-chapter book on Gen AI. Now you can just use low code, no code. And we pre-built these agents to do a whole variety of things. You can extend them yourself without programming them. You can add new things. We're actually going to show that in the keynote tomorrow, where we show an agent that solves 90% of a customer's problem but couldn't do one thing, and we go in with a couple clicks and we show we add one more action, a custom action, and then it can do 100% of what the customer wants.

Sure. Our view is that the LLMs are really the reasoning engine. But we have to certify them, and we have a trust boundary to make sure that everything that happens within our trust boundary is according to ethical use and it screens out bias. And so we take that part of our trust core value really seriously. But we do think LLMs over time are going to become more of a commodity. If you go to Hugging Face and look at the leaderboard, they're all hovering around in the mid-90% accuracies. They're going to get to the high 90s, and then at some point, the differences are going to be imperceptible to the naked eye: 98.7% accuracy versus 98.5%. This is like the old database wars with Codd and Date, like who had this many transactions per second. So we think the real value is in the data and the metadata. And we have advanced RAG techniques. We're doing something called ensemble RAG, which is beyond basic RAG, that really gets us far higher accuracy. And the final thing that we've done with our Data Cloud is we think what our customers want is to take structured data and unstructured data and put that all together in the output. And we think that's also a unique advantage that we have. We have all the conversations in Slack, we have threaded discussions within Salesforce, and we have all the structured data in Data Cloud and in all of our transactional apps.

Yeah, you just gave the answer: AI runs on data, and therefore the data is critical. If you watched our ads, we now have rotated to Agent Force ads, but we were running ads with Matthew McConaughey saying 'Data is the new gold.' We really believe that's the gold of the AI age. And it needs to be protected. We have our own backup and security solutions, but just like I said before, we have an open platform where we have ISVs like Own. It was originally called Own Backup; they have five products actually. The backup is their biggest product, but they do archiving, data security, data masking, which will augment our Shield products. And once the transaction is approved, it's not approved yet, not finalized yet, but once it's done, we think it's just one more way to reinforce we have five core values in the company, and the number one value is trust. So people trust putting their customer data in the cloud with Salesforce, and now that trust will just be that much furthered.

Sure. Because we add in we have 15 industry clouds. So digital transformation is an industry-specific thing. We add hundreds of data objects to these industry clouds. Maybe you care about the account balance or the credit score of someone in banking, or maybe you care about the billing profile and how many residents somebody has for the energy or the telecom or media business, so you turn on the service in the right location at the right time. And so each one of these industries has industry consumer trade promotions data in consumer goods companies. So each one has different data. We can use that data to inform the AI in the flow work because we've codified the best practices for sales, marketing, service, and all 15 of these industries. So we've taken advantage of that. But we also have the capability in Salesforce Research to build our own models. So we build a code gen product for our Apex programming language, which is a JavaScript variant. We've built a number of models, and we see a lot of action in the future in small models, in domain-specific models. And so we are working on some new ones in that area that we think are going to be very exciting.