From AI to ROI with Agent Fabric | MuleSoft Keynote, Dreamforce 2026

Please welcome Senior Vice President and General Manager, MuleSoft Salesforce, Andrew Comstock.

Good afternoon, good afternoon, good afternoon. I'm so excited to be here. Now, I also know that this is likely for most of you the last session, and we are the last thing between you and Dreamfest tonight. So, I'm so thankful that you've chosen to join us here in person. But even more than that, thank you to our incredible customers. Thank you to our absolutely amazing partners. And thank you to our world-class community. Now, for many longtime Muleys, this might actually be a slightly different, in a really positive way, Dreamforce for you, because we've been on a journey. And you've probably heard more about MuleSoft and its Dreamforce than you have ever heard before. But despite the new name with Agent Fabric, it is still at its core the same fundamentals that we've always been investing in. It's about agnostic platform and connecting your environments. As a result, integration and API management remain core to every single thing that we're doing. We want to just take a moment to highlight all of the incredible work that we've done bringing integration and API management forward into the agent age. I've been incredibly enthusiastic in hearing about talking to customers in the last few days about some of the things we've been doing, hearing about how customers are now rolling out into new countries as we mount our control plane and our runtime plane into new regions, helping them solve data sovereignty issues. Or even more, how we've done something that sounds very small, and we recently released a new agent to help you do runtime and Java upgrades fully agentically. This is the type of stuff we want to do to make our customers' lives just that little bit easier in the things that they've been doing for over 20 years. But we also know the next generation is coming, and Dreamforce last year talked about AI at scale. And I think here in San Francisco, every single billboard you walk by is basically screaming at you about AI. But despite that, we know that enterprises are still on the journey themselves. 94% of companies still report that they aren't seeing strong ROI from their AI investments. And we sit down and we talk to customers about why. It comes down to a lack of confidence and a lack of control. When we think about what the challenges they're running into, AI is simply everywhere. It's in your hyperscaler, it's in your apps, it's in your data platform. Heck, it's even coming from the frontier models themselves. This proliferation of technology is being strewn randomly across the average enterprise and causing straight chaos. Well, chaos isn't something that scales. And so as I spent time with more customers and start digging into what is challenging them, what is, being the barrier to their acceleration? We see 3 things. We see a lack of visibility and control and ensuring what agents are even doing, who owns them, where did they come from, what data can they access, and what are they even following the rules that you instructed for them. In addition, we've— you see the agents aren't taking meaningful action. You can make a really simple agent, upload a document, get simple answers, maybe a recommendation. But if your agents aren't able to take action in the systems that are meaningful to your business, they aren't having the return you want. And as a result, despite the investment, we are still seeing employees report that they're still stuck in working with the mundane actions. And lastly, they might be following all the rules, might be taking all the actions, but it's doing at a cost that's just totally unpredictable. AI costs are surging, but even more, it's the unpredictability where we're hearing teams are getting bills at the end of the month that are increased by 20%, 30%, 40%, and they don't know why. They don't know what team has been using them and what they're going about. As a result, we need to be thinking about this differently. So what do you need? You need trusted governance. You need consistent policies and guardrails that ensure you can control how your agents are going to behave. You want to make sure that you've got trusted agent outcomes and actions where that they can take tangible value in the actions that they're taking in your enterprise systems. And you need to make sure you're thinking about this with sustainable costs, making sure that you can predict and control how your agents are going to cost you each month. That is what we built Agent Fabric for. Agent Fabric is the control plane for trusted action and agent governance. It enables you to have discovery for AI across your entire enterprise landscape regardless of what platform you chose. It enables you to connect your agents to your enterprise system your business relies on and ensure that they're governed so you can scale agents more rapidly into production. And in addition, we can make sure that our costs are now scalable, sustainable, and predictable, that you have control of where the tokens are being spent by what team and where. In the market, it's called an AI control plane. And this AI control plane from Agent Fabric is the one platform you need to be able to manage agents regardless of the platform, the model, or the ecosystem that they're working with. But I'll also note, for those who go to a lot of keynotes, This is not just Agent Fabric in isolation. Agent Fabric and the AI control plane is the foundational piece to the enterprise AI harness that Salesforce just announced yesterday in Rohan's keynote. But if you don't believe me and you want to hear what customers are doing, we have some absolutely incredible customers who are going to be joining us on stage to tell you their story about ROI and AI. And to get us started with that, let's start with a video. So let's roll that film.

There was never just one lightbulb moment for us. It really was a storm of moments.

Capita is an AI-enabled services provider. We run some of the most complex processes to serve the public sector. the government, and regulated industries. We enable close to 20,000 employees with the ability to build agents.

We have agents on different platforms, agents coming from third-party vendors.

Imagine 20,000 times 5 agents, you're gonna have 100,000 agents.

There's agents everywhere.

We anticipated that this could lead to agent sprawl, which means you don't actually know what those agents are doing.

You can't govern what you can't see. And you can't scale what you can't govern. It's just this perfect, authentic storm.

That is what Salesforce helped us with. Millions of people rely on the services that Capita supports every day. We use AI, but at the front is a human. That's why Agent Fabric fit the bill perfectly.

Agent Fabric helps our assessors to get back to the part of the job they love, which is connecting with people.

Agent Fabric is essentially a control plane to control agent sprawl, allowing our employees to build anywhere and deploy anywhere with all of the security governance and trust layers in everything that we do.

So they're able to focus on the human interactions. Our business teams want to move really fast with AI. My team's job is to make sure that they can do that safely. Agent Fabric enables us to do both. Just 6 months ago, our executive team wanted to see how many have we got, where are they deployed, I couldn't answer those questions easily, but when I showed them Agent Fabric, they were stunned. With the Agentic technology, an assessor's day has really changed. It's gone from being able to do 1, maybe 2 assessments in a day, to 3, 4, even 5 in a day. In one of our client engagements, we've seen a reduction of admin overhead by 25%. In case management, 30 to 40%. 40% cost reduction. I'm really proud of the work that I do for Capita. That's the impact that we're really looking for.

When a new tool comes along that can drastically improve quality of service, you jump on it.

It was like going from agent storm into calm.

So it's a win-win. Why wouldn't you?

And and speaking of human interaction, hopefully you're paying attention. I want to continue the story live here by welcoming on stage Adolfo Hernandez, CEO of Capita. Thank you so much.

Yeah, thank you. Thanks for having us.

By the way, incredible video. Now you know. Capita has been embracing Agent Fabric for their operations. Why go all in instead of just dabbling or starting?

Well, I guess I was the one that asked Tina the difficult question, right, um, back, back in the day, and I remember that, uh, very vividly. Listen, when I, when I took on the responsibility to be, uh, to join the company as the CEO early in 2024, I did it with the belief that AI was going to fundamentally transform the whole business process services industry. Now, we didn't know anything about agents at the time, but we knew it was going to be fundamentally transformational. So when you are in that situation, you might as well go all in, right? You can't just dabble. Now, we needed to get tens of thousands of people, those process experts that we employ, to start thinking, to start innovating as to how they could improve the service we produce to citizens and, and than the people we serve. So that was the first challenge. But then they started innovating. They started creating all of these agents. They got into the hundreds of agents. Then we needed to make sure what agents on what guardrails, what security frameworks, when do they see the day of light, the light of day, when do they retire, what guardrails do we put in place. So my discussion with Tina was like, Tina, we're making a lot of progress, but how do we get this under control?

Absolutely. And, and part of that plays into control, plays into return. So how do you define return on AI at Capita?

Well, for us it's absolutely fundamental. You've got the financial returns and then you've got the operational returns and then you've got the customer return. So, so if you look at some of the operational numbers, well, we, we've, we've seen that 25% that was mentioned in the video. We've had some, uh, many engagements where we see up to 50% reduction of the average handling time. We're seeing it, the getting it right first time in the 90%. We're actually seeing customer satisfaction going up. And then financially, you're starting to see those savings flowing through the cost of delivery, which makes our offerings more competitive. You can see an improvement in the win ratio of the company. So, so it's fully circular. You start by getting the efficiencies, you redistribute that capability, and then you start winning more, which tops up the top line again.

I mean, that's clearly a payoff in financial results. Is there other things in the overall experience you're seeing improve?

Things are improving. I mean, you look at where we are in the last 6 months, we retained 100% of the customers that we have. We have a win rate in excess of 80%. Our customer net promoter score is the highest it's been since We started tracking it a decade ago. Our innovation is way better than it was a few years ago, so it's definitely producing. But I will say these investments are not isolated. You can't really just put money into AI and measure the return on AI. You put AI in as an ingredient of your total value proposition and you see a holistic return on, on everything that you're doing.

That's incredible. Well, please, thank you so much for joining us on stage. Thank you for having your team film the video. I really appreciate it. Can we give a round of applause Let's give a round of applause to Adolfo and the entire Capital Team.

Thank you.

So what we're going to talk about now, we're going to go deeper, and we're going to talk about the transformation and how you're going to go from AI to ROI. Now, to get us started, please welcome on stage Ashley Hathaway to talk about trusted governance.

Thanks, Andrew. If we were all to reflect back on the past year, most of us have been involved in some sort of AI initiative, either ourselves or with our teams. But how many of your agents do you actually trust to run in production? Trust to access your critical data? Trust to take action like one of your employees? That's why I love that line in the Capita video. I don't know if you caught it. You can't trust what you can't see. It's short, sweet. It's exactly to the point. Makes me think of a conversation I was having with a customer a couple weeks back. I was trying to get a sense of the scale of their operations. They told me they had about 600 agents, give or take. running across their enterprise. So we got them started with Agent Fabric. Do you know how many agents we found out they actually had? 5 times that many. They had over 3,000 agents running across Agent Core and Copilot. And those 2,000+ additional agents that weren't even on their radar, well, they weren't sure if they were accessing critical data, Or creating any sort of compliance risk. That's the reality. It doesn't matter if you don't know about agents. Your organizations are being held accountable for all of them. So if you get ROI from being able to trust your agents, first you need to actually be able to find them. And once you can see them, then you need to be able to control and govern them no matter where they're running. That is the premise of Agent Fabric, having one place to first be able to discover and catalog your agents automatically, then be able to control and govern them consistently across areas like data privacy, compliance, identity through the Omni Gateway, and then be able to swiftly intervene with a kill switch in case anything goes wrong. I firmly believe that governance, it's not actually here to slow us down. It's here to let us continue to do that experimentation, but then scale it quickly. So we're going to show you how. Throughout today's demos, we're going to work with Cedar and Company. They are a fictional home goods retailer, and like many organizations over the past year, they've given their teams flexibility to choose the agent building platform that works for their use cases. But this means they've got agents just scattered across numerous platforms now. And despite all of those agents that they're building, very few are actually having any real value. In fact, most haven't even made it into production. Their security teams lack the controls that they need to approve with confidence. So Cedar needed a way to preserve all that goodness, that innovation that was happening, but make sure they didn't introduce unmanaged risk. That is where Agent Fabric came in. So as we pull up the demo, I want to introduce our demo drivers today, Sue and Sonja. All right. We have said it starts with visibility. And we were here exactly a year ago announcing the Agent Registry. Here we see Cedar already has 250 assets in their registry. So, how did they reach that kind of scale? You all quickly gave us feedback that having a central place to see all your assets, that was great. But were you going to have to manually put them all in there? Fair point. So, in January, we launched Scanners.

Scanners.

And if you see this list of providers, this is what Cedar allows them to connect to the platforms where their teams are building, like Databricks, Amazon, Agentforce, and then automatically register those assets. If we take a look at Amazon, notice it's not just scanning for agents, it's also scanning for MCP servers. and APIs. That's because Cedar's AI landscape, it's many layers deep. So they value having that centralized visibility across the full tech stack. They've got an Agent Core scanner configured, so let's take a look at that. Notice just this morning it found 50 new agents. And later this month, Cedar is going to be able to connect into their custom agent services as well. That is the power of scanners. I think we've driven home visibility enough. So what additional value do you get once you have this visibility? Well, Cedar now can think about centralizing their governance strategy. And thanks to Agent Fabric's headless capabilities, teams can do that from the tools where they choose to work. For me, that's Claude. So, we're going to pop over there now, and I'm going to prompt it with a simple question. Which of my agents are high risk? And then recommend policies to start to remediate this. We've seen enough demos. Waiting for the response, it's not the exciting part. So, we're going to swap over to one where it's preloaded. We're going to start to scroll, and this is another view of the catalog of agents that we were talking about. So, we're going to scroll through that, And holy moly, those agents that we just scanned in, zero of them are compliant. We're not even sure if they have the basic security policies applied. This means critical data maybe is just passed willy-nilly. We're not too sure who even has access. But before I fret too much, if we keep scrolling, Agent Fabric has recommended 2 policies we can apply, one for authentication and one for PII detection. I love these proactive suggestions, but now I actually have to execute on this. Before, I would've had to go to these agents one by one to apply these policies. But now, with a single prompt, I can apply it to all these agents. But Sue, before you press enter, I think we can do one better. These are pretty basic policies, right? They should be applied across the board. So we're going to save this as an automated policy. What this means is that all these agents, but also all future agents, are going to get this by default. Default. This is that first building block for Cedar in building their universal governance strategy. And now their teams can start moving agents into production more quickly. So what's next? Let's chat roadmap. What was that? Oh, seems like Some sort of Slack notification came in? All right, something's up with the order agent. Sorry, roadmap's going to have to wait a second. Let's check the security page. They can click the link, and it's going to take us back into the Agent Fabric dashboard. And sure enough, the order agent has been flagged. So, we're going to drill in, try to see what's up. It looks like it's been caught in some sort of loop. Just spamming the same MCP tool over and over. Tokens going down the drain. This this happens. An agent can get stuck on a problem it's solved 99 times before. But there there's an emergency protocol. The security team has an agent kill switch, and it can come in and it can. quarantine this agent. This is going to immediately cut off access, and it's going to contain this rogue agent before it causes any more damage. This is going to give the teams time to come in, audit what's wrong, fix it, and when the time is right, they can come back and restore access. That, that's that emergency brake that the security team needs But when we think about brakes, we often think about cars, right? But just like with cars, brakes actually aren't there to help you go slowly, except for Grandma. They're there to help you go quickly, but safely. I don't think we have any more Slack notifications, I don't think, so let's actually chat roadmap now. Because we are continuing to innovate, and we're also continuing to invest in our partnerships. One of the things we're committed to is being an open platform, and one of the things that means is that you have existing security controls and platforms that are your in your ecosystem, and so you can take those guardrails that you're writing on other platforms: Amazon, F5, CrowdStrike. and apply them from a central control plane. You guessed it, Agent Fabric. Because you need to be able to govern your agents in order to trust them. And once you trust your agents, that's when you're ready to let them start take action and driving real business outcomes. So to talk more about agent outcomes, Please join me in welcoming to the stage, Simon Goddard.

Hello, Dreamforce. All right, let's address the elephant in the room. Tony Robbins is next door. Can you hear that? All right, I want Tony Robbins to know that we are motivated enough. Can I get a cheer from this room? That's what I like to hear. All right, where are we in this keynote? We have governed agents that we trust. Now we need them to do something valuable, right? That's the whole point of agents. And the beauty of agents is, well, the outcomes are limitless. There is so much we can do with agents, and who knows what we'll be able to do in 1 week's time, 6 months' time. So before we get into the technology part of agent outcomes, actually the most important thing we all have to do is a mental exercise. We actually have to decide before we put an agent into production what is the desired outcome. Now, you will not believe the irony. What I was about to do was to take us through a guided moment of meditation. I, I kid you not. Look, listen, look, we even have, we have a A little fun astro here and some music to take us through, and we're going to attempt to do a meditation, but we may be interrupted halfway through, so bear with me. What I'd like us to do then is to try and take a deep breath in through the nose and exhale, because we are going to center ourselves on agent outcomes. I want you to think about what was the last agent you tried to create. There we go, there's the music.

The zen.

Think about the last agent you tried to create. What did you want it to do? How successful was it? What could it have done better? And what stood in the way? Hold it. Think about it. All right, come back in the room.

Ah.

Open your eyes and think about that. What stood in the way of agent outcomes? And I'm curious if either of these 2 problems resonate, because as I talk to customers, there are 2 consistent themes getting in the way. Firstly, connectivity. The agent is too siloed. It can't work in the systems, the data, the workflows that we need every day. The second thing is consistency. An agent that works once but then quietly fails the next week, that's not the outcome we need either. Agent Fabric solves for both. And today's demo, we're going to show you 3 key features. Firstly, the thing that we know and love MuleSoft for, connecting across your enterprise, connecting to any— Yes, any app, system, or other agent. The second thing, being able to do it quickly by reusing existing integrations or APIs, but turning them into agent formats they prefer, like MCPs. And finally, it wouldn't be Dreamforce if we didn't debut something new. So wouldn't it be great if you could continuously evaluate how an agent is performing, both before you put it into production and then throughout the rest of its lifecycle? Well, That's why I'm really excited to announce coming later this year is agent evals to Agent Fabric. So let's show you in action. So let's go back to the orders agent that Ashley just showed us from Cedar. We now trust it, but let's see if it can act in the way we want. Sue, why don't we ask it about the availability of an iron pan? that Cedar has. So the agent's thinking about it, it's seeing if it has what it needs, and it's come back and said, I'm sorry, I don't have access to the inventory system at Cedar. Now, in theory, this is the right answer for that agent, but for me, it's the wrong outcome. So how can we fix this quickly? Well, I know that Cedar has an inventory API already that can access that system, track orders, change orders, etc. So why don't we start there? And if we go to Agent Fabric, I can easily find that API. In fact, tech stacks are diverse. If we zoom in, you'll notice this API is actually being scanned in and run on a third-party Kong gateway. I'm joking. We celebrate this. This is exactly what Agent Fabric is here to do. It's designed to bring in your tech stack no matter where it lives, so you can do something with it. So in theory, I could connect this API straight to my agent. But we know over the last year agents work a little differently. They prefer to work with MCP servers. And to recreate this API and the hundreds of others across Cedar would take me weeks, months potentially. Well, with Agent Fabric, I can do it in minutes. I can click Expose as MCP Server, and just like that, I can have a more agent-ready asset. And the reason I can go quickly is I can do multiple in one go. So Sue, I also know this agent needs to interact with the product catalog. So let's throw that in for good measure. And I can control the outcome of the agent by configuring exactly what it can do. So a bad outcome for me would be if I woke up one day and the agent had deleted something from my product catalog. I don't need that on a Friday morning. So let's deselect it to make sure the agent only achieves my desired goal. Now that's all I want to do here, so let's click Next and deploy the server. And now let's go back to the agent with this server connected. You don't need to see me connect the server to the agent, so here is one I created earlier. And Sue, let's ask it the same question and see if we get a different outcome. So the agent's thinking about it. And it's seeing if it has everything it needs, which—

voilà!

That is a much better outcome. It's giving me the stock, the description, and I could even go further and action if I needed to. But for now, we've solved connectivity quickly. So let's talk about consistency. Wouldn't it be great if I could give this agent a performance review similar to how my boss checks in every now and then to make sure I'm doing my job, which By the way, how is this keynote going? Exceeding expectations. Wouldn't it be great if I could do that with my agents? But to do that across hundreds of agents isn't scalable. That's the insight that is leading us to introduce agent evals to Agent Fabric. Take a look at this eval I run— I've ran on that agent, and what you'll see is it's generated dozens of automatic testing scenarios to this specific agent. So for example, it's an orders agent. It needs to be able to clarify vague questions. It also needs to be able to route chargebacks to a human. And if we click in, you'll see it's rated it based on criteria that I've set. So how correct, how relevant, how complete was its answer. Now, it's great to do evals and see what's working, but we really do evals to see what's not working. So if we head back, you'll see it actually failed 4 of the desired tasks for the agent. Agent Fabric is here to make this easy to fix. So if I head up to the AI summary, it's going to give me generative insight into how I can fix it. And the differentiator The reason you want to tie your evals to the same governance strategies that Ashley just talked through is because I can immediately do something about it. So if I click investigate right here from the same platform, I can apply policies that it's recommending to ensure a better outcome moving forward. That is the Agent Fabric difference. That is how you manage and govern your agentic enterprise. at scale. And this is just a taste of what's to come in Q4. As I mentioned, the benefit of having this tied to your governance platform is the kill switch. For example, if you notice an agent is drifting whilst in production, you can pause it, quarantine it until you get it back on track. Now, this chapter is all about outcomes, and I cannot think of a better person to talk about outcomes with Agent Fabric than our customers. So to welcome 2 very special customers to the stage, please put it together for EVP Data Foundations America, Matt Detroyer.

All right, pretty amazing, pretty amazing. Thank you so much for being here with us today. And as much as I'd love to hear from more of my colleagues here, I want to make sure we're not caught in an echo chamber. So really excited to welcome up Christy Fisher and Steve Seigrist from ADT and Newell. So many of you are interacting with Newell and ADT on a daily basis, whether you're an ADT customer for your home or your business like 6 million others, or you're interacting with the Sharpie, and I've seen a lot of them around the conference. So thank you guys so much for being here today. Super excited to have you.

Fantastic.

So, Christy, let's start with you. ADT protects millions. How do you scale AI while keeping life safety compliance and customer trust absolute?

Yes, it's a great question. And I always like to start by saying we have the great privilege of being able to help save lives for a living. It is absolutely vital to us that we think about responsibility and responsible AI as we're looking at our life safety. And that with all of the rapid advancements in technology, we are over 150 years old as an organization. And so as you can imagine, as I look around this room, that is a very complex and fragmented ecosystem. And so we're very early on in exploring Agent Fabric. But even what we've seen through the demos is exactly why we are indexing and looking very deeply into leveraging Agent Fabric. It helps us from a governance perspective and really keeps those core responsible AI principles at the forefront of everything that we are looking at from a life safety perspective.

Thank you, Christy. Super awesome. Steve Newell, some iconic brands— Sharpie, Yankee Candle. We've talked about a couple of them. How do you balance rapid tech deployment with protecting flawless customer relationships?

Yeah, thanks, Matt. I think one of the first things that I would call out is that we really don't see any tension between rapid deployment and moving fast and wanting to take care of the customer. We let the customer decision— that customer's pace is the pace at what we move. So when we look at our business, really we run everything through a filter, and you're going to hear me repeat it many times. I'll say, are we easy to do business with. That's what our customers are looking for. And so when you look at our network, we'll have 41 global manufacturing world-class sites, 15,000 trade customers, 39% of our business being done outside of the US. That's a lot of room for variation, a lot of room for variability. But when it really boils down to our customers want 3 simple things: they want their goods shipped on time, they want them to be complete, and if we did something wrong, did we resolve it quickly? So that's the lens that we look through, and how fast we can do that is really up to the technology. It's up to our people to understand the customer's position, what they're looking for. And AGIC Fabric really gives us that view, gives us that scale to do something that's a bit bolder and a bit more dynamic than what we've done in the past.

I love it. Such a great focus on customer experience at the Newell Brands. Christy, you've argued against sort of building on one particular platform. Agnostic is something that we've talked a lot about. How does agnostic control plane help you govern a diverse set of models effectively at ADT?

Ultimately, it helps provide that confidence that we're doing it in the way of which we are comfortable as an organization. So I would say that is front and center for us. We have— and I'm rounding down— over 500 different applications and platforms within our organization. And so with that comes a ton of complexities and an understanding of where are these agents running and what is actually happening. And most importantly, to the point that you were making, are these impacting our customers and end users in the way that we would want? And is it actually helping to achieve the outcomes that we set out to achieve? So that is why that centralized body and ownership and governance is so incredibly important to us at ADP.

Steve, you've warned us, and we've had conversations. We were talking at dinner last night, sort of AI for AI's sake, right? And I've talked a lot about random acts of AI. How does a centralized governance sort of platform keep your team safe without slowing down innovation with disconnected tools across the enterprise?

I'll throw out a philosophical statement that maybe I invented, I don't know, but I'm going to take credit for it. It just makes a lot of sense. And I'll say AI governance is what separates AI strategy from AI experiments. And so when I look at it in, in our camp, as I said, the number of brands, number of sites, number of customers, we could really need to watch out for pilots or small ideas that don't scale and just know that there's capabilities to harness this energy and to put this together. Because some of these ideas are great, but if you're creating them kind of a one-off, you create capability islands. And we want to do anything that separates us from creating something that's fragmented. We want more of a cohesive strategy in how we're pulling stuff together.

Kind of comes back to a consistent experience for your customers regardless of what brand or medium that you're meeting them on. Christy, you've said chasing ROI too early is a trap. We talked a little bit about this as we were prepping. How do you sell your stakeholders across the organization on the foundational work needed before you might be getting returns out of it?

Yeah, I would say it is so incredibly important to think about all of the different parts that actually help AI to be successful. We oftentimes want to focus on only the model output or one specific, I'll say, vanity metric, but that is actually contrary to what an enterprise technology or an enterprise team needs to be focused on. And so we really look at the dimensions of what success looks like, what those outcomes truly are. They span quality, efficiency, sentiment, experience. Like, there are so many that we actually look at, and whenever we're evaluating the choices that we need to make to scale something, but if you make those trade-off decisions too early, then you're actually going to prevent that bigger process redesign and outcome from being achieved.

I think everyone in the room can attest to, you know, having that kind of vision and challenge inside of their organization. Great insight. Well, we're going to wrap here. Steve, you've called the best customer experience no, no experience at all or no service at all, right? Best customer service is no service. How do you measure AI's value in creating seamless operation versus sort of traditional metrics in the call center or the experience environment?

Our experience is going to be really similar to Christy's. We've got data for everything, which is great. We make smart decisions off that data. We've run our business off it. But if I go back to, like, what am I doing to really help the customer? I mean, we— everyone in this room loves a customer. We get that. And I have the luxury of leading a customer service organization. So when I say the best service is no customer service, it just means that things worked. We didn't have to intervene. We didn't have to get involved. And when we query our customers, and we know this from 600,000 Salesforce cases every year, what do they want? We want to be seamless. We want to be something where they say, I can count on you. When I place an order, it goes through. You ship it on time. You ship it complete. You ship it accurate. So our measures are going to be a little bit different. Our measure, our real measure is going to be, are we eliminating friction somewhere in that process? Are we getting information from that customer that friction is being eliminated? And then through Agent Force again, we get a platform, a way of governance to see how this is coming together. I said Agentforce, didn't I? I meant AgentFabric. But of how it's been pulled off. And really what we want to say is that, like, quit doing things that might be a small experiment. Do things that are big experiments that are really going to change the game for our customers and separate us from our competitors or do well by our customers.

Well, listen, we appreciate you guys going all in, being on the journey with us. We love our customers. We love the journey that we're on. We're excited to see where we go together, and this is a great partnership. So give it up for Christy and Steve. All right, so to take us away on our last chapter of the MuleSoft keynote, welcome our COO of MuleSoft, Ayeong Ahn.

Thank you, Matt. All right, so we are one step closer to getting ROI. We discovered and governed our agents. We empowered them to take action across the enterprise. And last but not least, and I would say most importantly, we have sustainable cost through cost management. So as a COO, I'm constantly having to make trade-offs. And yes, agents are truly valuable and impactful, but I don't know if the scale of AI is worth it if the economics don't scale. How many of you agree? Exactly. Cost of AI is top of mind for many executives that I talk to because there's LLMs, MCPs, agents popping up everywhere, and AI costs are stacking up way beyond all of our expectations. And as the great philosopher Notorious B.I.G. said, mo' money, More problems. And he was absolutely right, because these problems, they're not easy to solve, and we don't always know how to get started. Let's talk about that. How do you get started? Well, first, you want to stop token counting manually. You want to do that automatically. And second, you want to be able to set spending limits so that people cannot exceed their budgets. And last but not least, you want to optimize your token usage so that you can ensure that your token cost matches the value it creates. Super easy, right? I know it's easier said than done, but that's exactly why Agent Fabric is here. Agent Fabric empowers you to understand, control, and optimize your AI spend across your organization so that you can scale AI with confidence. You guys want to see it in action?

All right.

So, are you ready? Okay. Ooh, this is Cedar and Co.'s AI bill for the month, and they are well over their budget. And looking at this, they're thinking, what happened? Did we have a surge in AI usage? Do we have more teams using? And honestly, they have no idea. And the fact of the matter is, this is just one LLM bill that they're getting for the rest of the month. So how do we prevent surprises like this from happening? Well, this is where Agent Fabric comes in. Agent Fabric gives Cedar and Co. a single pane of glass to understand their AI cost. So let's start with the basics, token counting. For this, you actually get a 2-for-1.

What?

Number one, if you're using Omni Gateway, you are ensuring that your traffic is more secure, but in addition, you also get automated token counting. So as you can see on the top left corner of this page, you have the total token usage breakdown by LLMs as well as breakdown by users.

Great.

So that's tokens, but understanding your token usage is not the same as understanding the cost. And for that, you don't want to use some generic token unit cost for all of your LLM. No, as we've done here, you want to be able to put in your organization's unique token cost so that when Agent Fabric does the magic of calculating the cost for you, it reflects accurately the spend for your organization. Okay, so you have the token cost. And tokens. Now we can set in some controls. To do that, let's go back to the FinOps dashboard. And here you'll notice these things called model wallets. Think of model wallets like a prepaid debit card for your organization. And the best thing about these is that you can actually set a spending limit that you cannot exceed, just like debit cards. So going back to the surprise bill that we saw earlier, Now, I can actually describe the specific LLM and set a spending limit. So for Claude, now I will never go over budget again. Great news, right? Thank you. Now, we've set in the controls, and I think we can start to have some fun now. You guys ready to save some money?

Yeah.

All right. So I see a great example here in front of me. There's an agent that's about to hit its limit. So why don't we click on that and see what's going on? If we actually filter for this agent, we'll notice something interesting. So this logistics agent, despite having access to multiple LLMs, is actually only using the most powerful, thus the most expensive model. But we can fix that. Let's set up a semantic route. Think of semantic routes as an intelligent traffic controller for your organization. And its number one job is to make sure that your most expensive model only gets used when absolutely necessary. So here we've set up a route so that for simple requests like product and inventory, it goes to the GPT mini model, which is very capable of handling this and is much more cost effective. And for something more complex, like forecasting or analytics, we're going to scale those up to the Opus model. And anything that doesn't fall into either of this category, we actually have a fallback mechanism with an LLM attached so nothing falls through the cracks. So an optimization like this is wonderful, not only because it's saving you cost, but also you're leaving capacity for your most powerful LLM to be able to address your business-critical requests that may come in. All right, so we're starting to save some money, but I think we can do more here. You guys want to save some more money?

Yeah.

Okay, so here, if we go back to the FinOps dashboard, you'll notice that we actually have a tab fully dedicated to driving optimizations. And here, Agent Fabric is giving me a bunch of different examples of optimizations that I can derive for my MCP servers, my agents, and my LLMs. Why don't we actually click on one as an example to see what it's recommending? Ah, it's recommending the semantic caching policy. And this is a policy you can use for your most frequently asked questions or requests. So rather than to have your LLM process those over and over again, burning a ton of tokens, we actually, actually can store it temporarily so that you can reduce latency, get faster responses, and reduce your token usage. And the cool part about this is if we zoom in, they're actually— Agent Fabric is estimating the real cost savings that this particular policy can derive for your organization. Pretty cool. But that's just one policy. for one proxy. Imagine, what if you can actually apply this policy across all of your proxies? And in fact, if we go to the top of the page, Agent Fabric has done that math for me. It's telling me that the estimated savings that can be derived for Cedar& Co. is over $16,000, and that's money right back to their bottom line. So more money, less problems. So you just saw how Agent Fabric helps you understand, control, and optimize your spend across your organization so that you can scale AI with confidence. But we're just getting started. There's a ton more coming for you. This is a huge area of investment for us, And we would love to hear from you, from your ideas. All right, we're almost at the end, and I would love to bring to stage CMO of MuleSoft, Karuna Mukherjee.

Well, thank you, Ayaan, and a huge round of applause for our demo drivers, Sue and Sonia. You know, and maybe a little bit of applause for our background musicians as well. So what you saw today was really a lot about how to move from AI to ROI, right? How do you get your agents to take trusted actions? How do you get your agents to really kind of go and deliver those business outcomes? And finally, how do you look at getting cost back from that value that you're putting in? Now, there's a lot that we saw here, a lot of innovation, a lot— you got a preview of capabilities that are coming. We really want you to look at Agent Fabric as your one single AI control plane, and we want to make it easier for you to get started. And so I'm really excited to announce our global early partner champion program for Agent Fabric. Now, here's what's special about these partners. They are specially skilled in Agent Fabric. They have the resources, they have the expertise. So reach out to them when you're in the campground, work with them, consult with them, really get started on building out your AI architecture that's going to transform your agentic journey. Now, there is one more group of people that I am super grateful for, and that is the MuleSoft community. They have, they have been on this journey with us. They have innovated with us. They are our customers, they are our partners. So really, really grateful to you. Now, every year we like to hear from one special community member. And so this year I am really excited to bring on stage with me Brad Ringer.

Hi, everyone.

A longtime MuleSoft advocate. He's a principal system engineer at AT&T. I mean, you are the heart and soul of the community, so thank you for being on stage with me.

Thank you, Karuna.

All right, so you've been at AT&T for a long time. You have been part of building that robust foundation for AT&T from integration, and now AT&T is starting their agentic journey. MuleSoft as the foundation, with Agent Fabric? Like, tell, tell us more about it.

Honestly, it's been really rewarding. 5 to 10 years ago, I mean, for me, agents weren't anywhere on my roadmap, right? I was focused on building out APIs infrastructure to support our 120 million customers, and for me, avoiding the 3 AM wake-up calls that come with that, right? Um, but really what's helped is we built the foundation for our agentic future with all the work we've done.

No, thank you so much. Now, you know, you mentioned on how AT&T has benefited from products like MuleSoft, but I want to focus on you as a community member and being part of this amazing community, which is like over 200,000 members. How has this helped you progress your career?

It's really been a game changer for both my team and myself. I mean, 6 years ago, I was in a MuleSoft meetup. I hadn't built a MuleSoft API, didn't know how to do that yet, and really the community has helped drive that. All the other ambassadors, mentors, and customers really have helped build that foundation for me. And, I mean, the other thing is the build-out that we've done, the work that everybody's done is really, I couldn't ask for more.

No. Brad, thank you so much. Thank you for being such a positive voice for MuleSoft, for Agent Fabric in the community. We truly appreciate you. And in honor of that, we would like to award you the Golden Hoodie Winner of the Year for MuleSoft.

Thank you so much.

Thank you.

Thank you.

So proud to be a Muley. Thank you.

Thank you so much. So here at Salesforce, we believe feedback is a gift. And I want to tell all the community members out there, please continue giving us those feedback. And the one thing you have been telling us is make it easier for us to get started. All right, I have exciting news, so maybe a few minutes. So you have told us, make it easier for us to get started. And so we have taken that feedback really seriously, and so really excited to announce our new pricing and packaging for MuleSoft and Agent Fabric. Now, this new pricing and packaging model is really about making it easier for you to scale. It is making it easier for you to get started. It has got built-in volume discounts, so the more you scale, the more you save. And most importantly, it is a single credit across all the capabilities in your portfolio. So like I said, feedback is critical. Please bring it on always. Now, we've told you how to get started. We've told you about resources. But the one thing that always takes top of mind for us is innovation. So I want you to take out your phones, take a picture of this. This is just a preview of what's coming this year. Now, today you saw the Omni Gateway. We released that in May. You saw the cost management capabilities that we released last month. You saw the agent evals in one of the demos today, and we are just getting started. As an example, the agent eval capabilities will be extended to support AI gateways, MCPs, and LLMs. Now, you've heard a lot about the innovation today. I am going to give you some homework. I want you to ask 3 critical business questions as you go back. Think about the agents in your organization. Are they governed? Think about, are the agents delivering value? Are they acting in a trusted fashion? And what is the agent costing you? Now, look at these 3 questions. The one that is most important to you, head to the campground, talk to our subject matter experts, talk to our partners. And while you are there, go ahead and collect some trading cards. This is a collector's pack of MuleSoft and Agent Fabric trading cards. Collect those trading cards and then head over to Dreamforce— Dreamfest, rather. Enjoy Gwen Stefani, Usher, Thank you for being here. Have a great Dreamforce.