Dreamforce 2025 | Nancy Xu, Salesforce

Clips
More from Dreamforce 2025

Nancy Xu

VP, Agentforce

Salesforce

play_circle_outline Introduction of Nancy Xu, VP of Product for Agentforce at Dreamforce 2025.

play_circle_outline DON'T USE - Empowering Businesses: How Agentforce Enables 12,000+ Customers to Build Agents Without Technical Skills

play_circle_outline Enhancing Enterprise Operations: The Role of Predictable Agents and Advanced Scripting Techniques in Streamlining Performance

play_circle_outline Emphasis on business leader involvement for successful agent use case development.

Info
Transcript

Nancy Xu, Salesforce

Nancy Xu

VP, Agentforce Salesforce

In this Dreamforce conversation, Nancy Xu, vice president of product for Agentforce at Salesforce, joins theCUBE’s John Furrier and George Gilbert to map how the “agentic enterprise” is moving from demos to dependable production. Xu shares real-world traction: Agentforce now serves over 12,000 customers, with outcomes such as Adecco handling 50% of off-hours candidate inquiries, OpenTable achieving ~70% case deflection for dining questions, and 1-800Accountant deflecting ~90% of peak-season customer inquiries. She explains why ‘not all agents are created equa... Read more

explore Keep Exploring

What is the context and participants of the discussion taking place at Dreamforce 2025? add

What unique features does the Agentforce platform offer to customers? add

What are the key factors that contribute to the effectiveness and trustworthiness of agents in handling customer inquiries at enterprise scale? add

What are the key points regarding the importance of business leader involvement and ease of use for non-technical individuals in the context of enterprise transformation and the Salesforce Agentforce product? add

bolt Powered by CUBE AI

Nancy Xu, Salesforce

search

>> Welcome back. Welcome back, everyone. I'm John Furrier, host of theCUBE here, live in San Francisco, California for theCUBE coverage of Dreamforce 2025. Joined by George Gilbert, our lead analyst on data platforms and Salesforce. Got a great guest, the VP of Product of Agentforce, Nancy Xu is here. And we're going to break down the product. We're going to drill down on what does it mean, what's some of the successes. Now, the second year as out in the open, in the wild. Nancy, thanks for coming on theCUBE, and appreciate your time. Congratulations on having great products.

Nancy Xu

>> Thank you, John. Thank you, George. Thanks so much for having me.

>> So Agentforce, Dreamforce, probably be called Agentforce at some point because it's Agentforce everywhere kind of vibe here, but super powerful platform that's emerged over multiple years of work built from the ground up we heard. Really enabling the agentic era, and you guys have a lot of data. Your customer data, Slack has data, just everything is kind of coming together and Agentforce is the center of it. You have the keys to the Kingdom. Tell us about what's going on in Agentforce, why is it so popular right now and what are some of the key things that's jumping off the page?

Nancy Xu

>> Absolutely. And John, we shared so many great statistics earlier this morning, but we have over 12,000 customers on Agentforce today, and we're at a point right now where it's so exciting because customers are bringing their data into Agentforce, right? And it's not just about them bringing their data into Agentforce, we have the tooling layer that we've already built for them and our customers are now able to unlock a new level of agentic enterprise that wasn't possible before. And I think part of this is not only are we making it easy for our customers to build agents, we're making it easy for anyone in their organization to build an agent, right? I think there's so many great agent platforms out there, but I think what's critically unique about the Salesforce and Agentforce platform is that we make it possible with the platform interface for your non-technical people, your admins, your developers, your service teams to participate in that agent building process. And this for us is super critical based on our learnings for customers to be successful.

>> One of the things we've been hearing and observing here at Dreamforce in the industry is the hype cycle. It's going from proof of concept, nice demo to reality so that everyone wants to know, "Where's the proof? Show me the agents." And so that brings up the question of where the use cases are kind of shining, what's winning, kind of what's struggling or what's lagging behind. Could you share your thoughts on some of the impact? Because it's a show me the money, show me the agents. Where's the traction? What's winning and what's struggling out there? Because you need scale, governance and whatnot, so what's your thoughts on that?

Nancy Xu

>> We have a lot of customers across sales service, of course, but also a lot of other industries beyond sales and service or functions. Adecco is one of our customers and an amazing partner with us. We help them handle 50% of their candidate inquiries that happen during off hours. We also have OpenTable, consumer company. We handle about 70% of their case deflections for them today when people ask about dining questions. We have other customers like 1-800Accoutant. I like this because it's a great example of how scalable agents can be. During peak tax season, 1-800Accoutant, we help them deflect 90% of their customer inquiries. And that is so phenomenal because especially for a company that is so driven by cyclical events, we can help agents scale up and scale down based on what your organizations actually need.

>> What are some of the things that people don't understand about agents? Because agents sound very futuristic. Here, fleet of agents. What are some of the misunderstandings and the realities of how do you deploy this at scale? Obviously, we know what chatbots do, everyone can see that. That's not really a use case. But as people start figuring this out, they start to see visibility into the economic value, the productivity value. What's the biggest things that are misunderstood or things that people should know about the agentic movement?

Nancy Xu

>> Not all agents are created equal, John. I think what we see now is the word agent is quite overloaded. There's a spectrum of agents in terms of how simple the task they can handle all the way to complex workflows. And what's critical is that what we found is for customers to be successful, they don't just need the core LLM capabilities that make your agent powerful and adaptive, but you also need to have predictability and control for your agents for them to be trustworthy at enterprise scale. I'll give the 1-800Accoutant an example. You need to be able to handle hundreds of thousands, millions of inquiries predictably every single time according to your business processes, and we're seeing that this is a big unlock for companies that can do this. It's why we introduced Agent Script this year at Dreamforce. Agent Script is a brand new scripting language that we built that enables our customers to have those predictable agents.

>> The thing that I was asked at a dinner last year at AWS re:Invent was how far in the future we'll be able to book a flight, pick a vacation, book a flight, pay for it and not worry. And I'm like, "Oh, at least five years." But now, that's kind of accelerated. So if I'm going to go on vacation, I don't yet know where I want to go, so I'm kind like maybe creative, non-deterministic what do I want to do. And then once I decide, I then have to go to a payment rail. This is kind of where agents kind of happen. I can see Agentforce enabling that. Take me through your mindset on the product side. What has to happen? What's the sequence of events that the infrastructure has to enable? Because that's going to invoke maybe multiple agents, one to handle the payment rails.

Nancy Xu

>> Absolutely.

>> It's very specific. One to handle the generative side of it, where do I go. Once that happens, it kicks off a progression.

Nancy Xu

>> That's a use case we literally do for our customers today. We have a customer Finnair where we help them handle many of their travel inquiries on their website today, and so it's a great example. When you think about that particular use case, it's all about the tools we unlock for our customers to actually be able to bring in their workflows into Agentforce. And then of course, we have capabilities like MCP and A2A, which become incredibly powerful as you start to think about multiple agent frameworks. So in your example, you might have an agent that's just in charge of your payments. There might be a different group of people building that agent. With A2A and some of our internal Agent2Agent work, you can now have your agents call other sub-agents within the same framework, and that becomes really powerful. And then, of course, there's the tooling in the MCP layer. Part of why Salesforce is so powerful is we have customers that have built millions of flows inside Salesforce. And these flows, you can now just take them and bring them directly into your Agentforce workflows, and that is really powerful. So it's about unlocking the full capabilities of the Salesforce ecosystem for our customers. And then going back to my point earlier, we make it possible for those business line leaders and those subject matter experts to participate in that agent building process. Because right now, a lot of people talk about the foundation models, but it's not just the foundation models that are important. In fact, I would argue that we're at a point where the reasoning models are pretty good, right? They're going to get better, sure. But the reasoning models are pretty good. The challenge right now is getting the context in, and the context lies oftentimes not with the really technical developers on your team, but those business leaders, those operations managers and Salesforce is uniquely positioned to help bring their information into the ecosystem.

George Gilbert

>> So let me build on that, because what we were talking about before is there's like thousands of agent development tool startups in addition to all the incumbents. I want to hear you riff on the secret sauce that you just touched on. It's not just the context and the flows as tools. It's the entire Salesforce platform is like Data 360. If an agent is like a bot, it helps it perceive the state of the business or the customer. And then all those flows or workflow invocations, they are now native tools. And then the script lets you compile something that has partly deterministic things that have to happen and then partly things that are adaptive. So the whole package is something that it seems to me no one else can offer if they want to deal with customer related data end processes. Help us understand that, and that's only the design time, then operations, it goes even further.

Nancy Xu

>> Absolutely. We have trillions of records in data cloud today. Customers have a lot of their data there. It's our job to help them then make that data actionable for their agents. We've invested a lot in the Salesforce metadata schema, and you can think about this. Salesforce has spent more than a decade figuring out the metadata framework that enables all of our customers, everything from the most complex enterprise use cases to SMB use cases be able to represent their organizational workflows and frameworks. That's really powerful in and of itself. And we pull that all in into the Agentforce ecosystem, which now when you combine it with these workflows or flows, enables you to build these end-to-end agents. And of course Data 360, it's not just data you have in Salesforce ecosystem. You can zero copy your data, you can bring your data from Snowflake if you want into the ecosystem. And then once you have all these agents, it's all about observability orchestration of those agents.

George Gilbert

>> Right. The day two stuff.

Nancy Xu

>> Exactly.

George Gilbert

>> Which was like I'm in production, now make it simple for me to do my evals, to learn from those, that sort of thing. Help us bucket the types of customers where they are on that journey, where they are trying to get their data in order, trying to harmonize all the flows, like clean up the metadata, where are they? Give the audience, as Salesforce customers, a maturity model to think about. Do you follow me?

Nancy Xu

>> I think about the maturity model from a customer perspective as actually being quite homomorphic or mapping very closely to the model that we have for what the agent development life cycle should look like. In the first stage, it's the customers that are in what I call ideation. They're trying to figure out, get their data in order, figure out what the use cases that are going to be the most impactful for their business are. What does success look like for the agents? What are the outcomes they want to measure? Stage two is really let's get our first agent deployed. And then stage three is I have my first agent deployed. How do I observe the behavior of the agent? And not only that, but how do I take the behavior that I observe customers using for my agent and actually use that to improve my agent over time? And hopefully, over time, we make that a self-learning process too. We haven't gone there. We're working on it. The first point, right? So many use cases, we can talk all day about how we can do enterprise transformation, but the core thing is business leaders need to be bought in, understand what the use cases are really trying to drive and measure the outcomes. The second stage building, we are going to make it so easy for anybody to build an agent. And you asked about what's so special about the Salesforce Agentforce product, right? This is part of it. There's no other product that makes it easy for those business leaders and the non-technical people to build those agents with experience so that anybody in your organization can participate in that process.

George Gilbert

>> Maybe drill down on that part. I don't know if we didn't get enough detail yet or I haven't seen enough on script, and then Agent Builder where you can work in low-code, but I think it generates pro-code so that, in other words, you have the accessibility for the business user, but the control for the pro dev type person. And I assume that, that's going to be critical for building really more sophisticated agents.

Nancy Xu

>> Absolutely. So underneath the new Agent Builder is Agent Script. So Agent Script, you can think of it as the programming language that powers the agents. If you are a developer, you can go straight to Agent Script in the new builder and you can author in that language. We support conditionals, we support all kinds of complex logical use cases. Now, if you're someone who maybe is less technical, you can go one level higher, you can write directly a natural language. It literally looks kind of like a Google Doc or any kind of document editor. And you can write a natural language, describe what you want your agent to do, and then what that will do is we'll then compile it to Agent Script for you. There's also one level, even higher. You can also vibe code your agent by talking conversationally to it. So we have a conversational agent, kind of like what you might experience in a product like Cursor today, but for building agents, and I think this is going to be so game changing, right? Just think about it. Anybody can just describe the agent they want, get that agent built for them. And then critically, when that agent is built for them, they can understand what the heck was built for them, because it's now described in natural language. You see this with coding agents, you can tell a coding agent, "Hey, I want this application to build me XYZ thing." But then you look at the code and you have no idea what it just did for you. So you can't really riff on top of it and continue to iterate.

George Gilbert

>> So is the vibe coding agent is a separate sort of branch for agent building from Agent Builder and script, those are two separate?

Nancy Xu

>> They are one unified experience.

George Gilbert

>> Okay, one. So you can go from natural language to script to vibe coding-

Nancy Xu

>> Yeah, you can talk to the chat and say, "Hey, build me an agent and I need a topic for order refunds." And in the live experience, you will see it author for you the actual diff in your canvas of what that should look like.

George Gilbert

>> Okay, that's pretty powerful.

>> I know you got a hard stop. Final question for you, adoption. Is it unlocking more momentum? What's your forecast on enterprise adoption? Governance is coming to Informatica. It's the scale trajectory going?

Nancy Xu

>> We're super excited. We see a lot of momentum for our agents today, especially with the new Agentforce 360. We're going to make it possible for more use cases to be built and for customers to get to that next level of scale that they're looking for with more predictability.

>> Nancy, thank you for coming on theCUBE. We'll have to continue this another time. We'll certainly-

Nancy Xu

>> Absolutely....

>> give you a call back in. I know you got a hard stop. Thank you for coming in.

Nancy Xu

>> Absolutely.

>> VP of the product, Agentforce. The center of the action here at Dreamforce. Agentforce is the future of this company as people are going to be programming their agents with natural language, vibe coding. This is going to set the stage for having agents work on our behalf, make us more productive. Of course, we're doing our part bringing that data to you. Thanks for watching.