Speaker 1: 0:22

Welcome. We're here at Salesforce TDX 2025. We're talking to Andrew Comstock. He's Senior Vice President and General Manager of MuleSoft, the iPass solution of Salesforce. It is yes. If we start with the wall of MuleSoft, did the strategy and wall change over the years? Within MuleSoft it looks like it's more positioned to serve Salesforce than it is to be a dedicated iPaaS solution.

Speaker 2: 0:49

Yeah, and I think I also say so. First, we haven't. Our DNA really hasn't changed right. We are still very focused on being the general provider of capabilities. So, in fact, actually, salesforce is even isn't even our number one solution that we integrate to by far and away the most common. It's very high, it's, I think, number two.

Speaker 2: 1:10

But we actually see other ecosystems being the most commonly used by our customers and what we see that is what's really changing is our ability to work together and be effective. But we would just sort of note that the most integrated application in the world SaaS cloud application is Salesforce. So our natural affinity to Salesforce has been before the acquisition and continues after the acquisition, and what you're seeing now is a reflection of the sort of switch from the digital transformation imperative that enterprises were under to an agentic transformation and with agent force really leading the charge in agent and agent deployments. There's a very natural affinity for us to do that, even as we continue to remain neutral. We're looking at supporting as many of these ecosystems as possible and making sure our customers feel fully supported by MuleSoft in all the ways that we possibly can.

Speaker 1: 1:58

We're here at TDX. There were a lot of announcements around AI and agent force, like you just mentioned, and in some of those MuleSoft plays a major role. I think MuleSoft for agent force is the biggest one.

Speaker 2: 2:12

Yep, I'll logic Does that sound like that? You're right on.

Speaker 1: 2:15

Yeah, can you explain a bit what it does? Yeah, yeah.

Speaker 2: 2:21

So MuleSoft for agent force is sort of an encapsulation of a couple of different components really designed to optimize and help customers get deployed on agents right and bring existing capabilities that they have what we have from MuleSoft, to make them more agent friendly. So the two of the headliner capabilities within that are what we call topic center or API catalog, and there's very similar concepts. Topic center is about empowerment, making sure that developers can make their APIs ready for agents by adding agent instruction, a key element to make your APIs more effective for agent actions. Api Catalog can do the same from the Salesforce environment and can unify across the entire Salesforce platform to make those accessible. So for the first time a Salesforce user can see APIs across Heroku, mulesoft, salesforce and the Salesforce platform to make those accessible. So for the first time a Salesforce user can see APIs across Heroku, mulesoft, salesforce and the Salesforce platform simultaneous in one place and with a single click deploy them as agent actions.

Speaker 2: 3:15

And again, that deploy as agent actions is sort of a nebulous concept that I'll ground for a second. An API needs an agent needs to know how to access the API so our two developers might talk and explain and ground how that API works. These instructions are the guidance for the agent to know how that API can work, and so this is a great way to power up your agents with third party actions across your entire enterprise ecosystem. And so we see this as we start making more advanced agents or we start discussions for headless agents being a critical capability to get access to more and more advanced actions. And we see MuleSoft our ability to power that with tens of thousands of customers and hundreds of thousands of APIs and actions that we can make being an incredible unlock for more advanced agents.

Speaker 1: 4:02

That's a lot. Trying to grasp a bit for the listener on how to look at this, because Fair.

Speaker 2: 4:07

I'm too excited. Grasp a bit for the listener on how to look at this Fair. I'm too excited. I'm too excited about the new capabilities. That's definitely my challenge.

Speaker 1: 4:14

Maybe we should split it up a bit because, from where MuleSoft comes from, it's creating API connections. Yes, absolutely, that can bring in data, but it can also execute actions for an agent.

Speaker 2: 4:28

In this case, yeah, so thank you for that, because what MuleSoft's heritage was was building what we call a business process in the format of an API and this think about this for people who don't know us this is about connecting two systems together. So it's you can be things like I want to integrate with Salesforce, with my ERP systems like SAP, or I could bring my workday data, integrate it with another employee tracking system and bringing those things together. That's really what's defined from any of our customers the business processes that they have. Right, this is the behind the scenes pipes that are powering a lot of the enterprises.

Speaker 1: 5:03

It's a connection in the back end that makes two separate software solutions work together, exactly, exactly. And on top of that now you created agents Correct.

Speaker 2: 5:12

And so now, thinking about that on top of that, with an agent, what we're going to say is agents want to take those same actions. You're going to want to be able to ask an agent when we said, for instance, a common example is like, book me a reservation. Well, how do you book a reservation? First, you need to think about what location I want, what specific maybe hotel room I would like to do, what dates I would like to have. All of those systems behind the scenes are APIs, and we've created reservation APIs for our hotel providers in the past. What we want to make sure those do is make them available, not just to systems that would access the API, but to those agents, and so that's what we're really going to power up is that final mile step, so that the APIs you've already built are now agent-ready.

Speaker 1: 6:03

Okay, is it fair to say that MuleSoft is also kind of the data fabric to?

Speaker 2: 6:07

the Salesforce data cloud. That's a great first great question I'm going to. I'll go back and challenge you. There's been a lot of definitions of what a data fabric is yeah, I know.

Speaker 1: 6:15

that's why I asked.

Speaker 2: 6:17

Okay, that's a great. So maybe I'll preface by saying to me what a data fabric is is thinking through the disparate systems an enterprise buys. The average enterprise today is buying almost a thousand different applications, both in the cloud, on premise, and deployed to hyperscalers.

Speaker 1: 6:33

Yeah, so in the worst case they have a thousand data silos.

Speaker 2: 6:36

Yes, and they live in a half thousand data silos and the data fabric has always been about how do I sort of stretch something across those and in the traditional world, mulesoft has been a great way to stitch all that stuff together, to bring all those things and exactly what you said of data fabric, what we've seen with Data Cloud do is bring it to another level of that by bringing actually into the data itself and help you standardize data and do things like profile reconciliation, where you can actually standardize different systems that have different information on an individual and help you standardize all that together.

Speaker 2: 7:08

So the way that we think about it is that data cloud is helping you standardize across your enterprise and that MuleSoft also helps both data cloud, bring that data in but then basically extend that into the rest of the world and extend that beyond just what you might do from Salesforce. So you can work in both and this is going to push your maybe limit. You can work in both sort of the and this is going to push your maybe limit. You can work in both the data world and a metadata world and being able to help standardize that information.

Speaker 1: 7:31

Yeah, because the data and the metadata creates the context for AI to operate. Yep, absolutely so. The data cloud can connect to some data sources by itself I believe Snowflake and Databricks and some others. Yep, and if you have these, let's say, legacy sources or data sensors out in the field that are not easy to connect to, then you use MuleSoft to leverage the data into the data cloud.

Speaker 2: 8:00

Yeah, exactly, and I think what data cloud? One of the really revolutionary things about what data cloud did is that they brought in what they call zero copy and so working with the Snowflakes and the Databricks the other example and many other examples that we didn't mention the snowflakes and the Databricks the other example and many other examples that we didn't mention.

Speaker 2: 8:12

They can bring in data without sort of bringing in the data and work with the information in this place, where it is, and so that's a way to connect this information without actually having to move it, and moving the data is a very expensive, very costly way to operate, and so, in not moving it but allowing you to work with it, is a really revolutionary approach. But the challenge is that zero copy works well, it doesn't work for everything yet, and so, as you want to start saying, how do I get to the rest of those and not just legacy applications, and it could be things like mainframes, which are still running around in the wild out there?

Speaker 1: 8:44

But you can also be A nice cobalt mainframe at a bank or something.

Speaker 2: 8:47

Yeah, exactly, and MuleSoft can help you bring that data into Data Cloud. But it could also just be applications. You know one of the things about the thousand applications. Some of those thousand applications in your average enterprise are really important and some of them are pretty specific, and MuleSoft can help you get into even those specific applications and make sure that we can break down that silo and bring it into Data Cloud or work with it in any way you see fit.

Speaker 1: 9:08

If we move back a bit to MuleSoft and the logic part of it within MuleSoft, when you created those API connections, you can also build some logic things with if then and else. You know if you kind of the route you should take or define which route you should take in certain conditions. I saw in the Salesforce workflow builder, which I think is based on part of MuleSoft technology, you can now replace all those if then else rules and put an agent in the middle and let the agent decide which route to take based on the event information coming in. So I was wondering can you also put an agent in the ePaaS module of MuleSoft already or is that still to come? You can you can.

Speaker 2: 9:51

Because, they didn't talk about it here. Yeah, yeah, so as part of our announcement, I mentioned that there are a couple of big pillars. One of the ones I hadn't mentioned yet was actually our agent force connectivity, and alongside our agent force connector we also have an additional capability called AI chain, which does connectivity to other agent and vector database vendors to be able to help drive some of those solutions. But you can actually, inside of Mule application today, replace components of your business logic with agentics based steps, and so what we're seeing is a lot of adoption and interest here, because where customers aren't ready to replace their entire system with an agent, but they are very interested in replacing some of that logic and so, in essence, they've got business processes they would build, but as opposed to coding it themselves, they're using agents to replace that and that prompt is, in essence, becoming their code. Their prompt is becoming the logic that they've implemented. So, yeah, you can do that today with MuleSoft as well.

Speaker 1: 10:45

And is it just as good? Or is there a higher fault margin? Because then you need the agent to interpret an event in a wide way Before you had it really hard coded. If an event is A, you run scenario B, and now it has to determine by itself which scenario it should run.

Speaker 2: 11:06

Yeah, and what I would say is, in a simplistic case where you're saying if it's A, then do this, if it's B then, do this.

Speaker 2: 11:13

You don't really need an agent for that, right, because that's actually not that hard to develop. What we're finding is that actually. What you're finding is people are saying, hey, I want to know, I want to route maybe a good customer. Well, what's a good customer? A good customer might be one with high status. It might be one who spends a lot of money. It might be someone who's brand new, that you can project could spend lots of money, right, like there's lots of potential definitions with that.

Speaker 2: 11:38

And where we see agents being really interesting is you could actually say, hey, here's the information I have on my customers, here's what my customers look like. Tell me if this is a. Tell me if you project that this might be a great customer. And having the agent start doing some of the logic and reasoning gives you more flexibility to do things maybe potentially a little bit differently than you did. The routing logic you're talking about can get challenging to describe all the potential scenarios and actually asking agents to help bring some logic and some reasoning to that is a way to potentially unlock new use cases that we couldn't do before.

Speaker 1: 12:10

Okay, that's interesting. I also wondered observability has also been a large part of MuleSoft keeping an overview of all those connections? Because what you just said, a large enterprise has near a thousand applications on average and it wants to connect them together. And then the e-pass came along as it may and said to us put everything to us so you can have an overview of what's happening and we will give you insights. Because if you have direct connections between all the applications, then you have to build them yourself. You also have to maintain them and that's a lot of work and it's pretty hard. But now I think organizations were used to managing those connections that they already were using, but now you're adding agents on top. Does that change the way you observe the APIs and how they're being used, or is it kind of the same, but just a bit more?

Speaker 2: 13:10

Yeah, so I think the simplest answer is it's the same, but just a bit more. Yeah, so I think the simplest answer is it's the same, but a little bit more, but I think the bigger answer, I think the reason that's the heart of your question, is that it's actually even more important. The technique that we would do, the technology, isn't necessarily something different. We still want tracing, we still want logs, we still want our metrics that drive observability, but what we want to do is we want to make sure that you can generate this across the entire ecosystem, because it's even more critical.

Speaker 2: 13:36

In up until sort of LLMs, computer technology was always the same. It was deterministic, you would say, as you said, if A, then B, then C, and it was always sort of operated the same. Agents don't always operate the same. So the criticality of unified observability is really important, and this is one of the things that we're so excited about being part of a deeply unified platform across the Salesforce ecosystem is that we're going to be able to bring experiences where you can go from a prompt inside of an agent to the agent, how it processed, to the agent, actions it took, and you can actually trace all the way down the scope and that's going to be something that we think we can do very uniquely.

Speaker 2: 14:13

That's going to provide incredible value to our customers in a fundamentally different way, because that agent is going to be different and that's going to be something new. It's something we haven't accounted for in sort of our architectures and our designs. We haven't seen our customers do, but we're seeing customers like Adeko lean into this as a way to unlock new business value, to build new products and new offerings. They're a talent management company and they're reimagining how they think about their recruiting and support functions, and part of that is making sure that they can do that effectively, because the customer doesn't know that they're using an agent right. That's the whole sort of value that you're bringing new solutions to market, and they're centralizing that across 40 systems. So how do you do tracing across all those, just as you said, and it's becoming more and more critical and more and more important in an agentic world.

Speaker 1: 14:59

You're the general manager of MuleSoft and Salesforce has this AI agenda. That's pushing really hard. Are they challenging MuleSoft to do things? Do you have challenges to cope with with the speed and the things they require to facilitate them?

Speaker 2: 15:18

So I actually look at this much more as an opportunity. And that's a silly thing to say and I know it's like when you get interviewed, you always say what's your biggest weakness and you always want it to be a strength but I'm sort of flipping this on its face, I mean.

Speaker 2: 15:30

I fully admit, but what I would the reason I say it's an opportunity is that we're at the cusp of something really big and these technology shifts don't happen often. Right, we always talk about, oh, it's something new and different, but the last time this really happened in our space was about a decade ago with cloud and the digital transformation. So, as we see this shift in here, we want to lead the way for our customers and again, I see it as a massive opportunity that we're part of the same family as AgentForce. We can meet and work together and collaborate on building better solutions for customers as part of our deeply unified platform. So I'm incredibly enthused about what we think we can do and where we're going to go, because I think that this is going to unlock the next generation, not just of how agents are going to work, but actually the next generation of an iPaaS solution is going to be agentic and being part of a leading provider, of that's making NIC in a better position.

Speaker 1: 16:23

But there must be a challenge, right? Not everything goes as easy as it sounds.

Speaker 2: 16:28

No, of course not. I mean, maybe I would joke and say that if it was easy, then anybody could do it.

Speaker 1: 16:34

Yeah, that's true, but I wonder, for example let me give an example you have these agents now that work API-based on a process, and that's fine because they're not tied to latency, for example. But if you have customers interacting with a bot, you have latency and I can imagine they want to get the latency down as much as possible. And if you're connecting to a bunch of APIs, you can add a lot of latency. They could. Yeah, is that something you're working on now to see if you can cut that down even further?

Speaker 2: 17:09

Absolutely, and that's where maybe I'll go back to as a new technology. I think there's going to be constantly evolutions, but what I'm so excited about is that we've seen customers bring these to production and then give us the feedback right, because that's a solvable problem. If it's latency is the problem or accuracy is the problem, those are things to solve that are fun to solve because customers are pushing us to them. It's much harder for us to theorize where we think the world's going to go and then pick the solution right, but I think agents have shown us sort of put the line on the sand of where the world's going, and it allows us to then build around it. So, yeah, again, I think you're talking to a group of people who are relishing the change right.

Speaker 2: 17:52

Mulesoft did this 15 years ago, 10 years ago, in leading the charge in the digital transformation and helping modernize an iPass. That was the core of what drove our business as we built over the last decade right, and even when Salesforce acquired us six years ago, we've grown from I think at the time it was roughly $300 million to a multi-billion dollar enterprise. We see this opportunity for the agentech world as even helping us to continue to accelerate. So we see this as an opportunity for us to work together, and a true one plus one equals three opportunity. And yeah, there's things we're going gonna have to do differently along the way, but it is something we're leaning into aggressively.

Speaker 1: 18:32

For the listeners. Then the last question you had some big announcements here, yeah. Is there more to come in a short run that they have to look out for from MuleSoft?

Speaker 2: 18:41

The answer is absolutely right. We had some we do some great announcements with these big events. You're going to see some incredible big announcements with Dreamforce along the way, but we have-.

Speaker 1: 18:50

You already know that there's a big one coming for MuleSoft. There's some big ones coming. Okay, good to know.

Speaker 2: 18:55

And but I'll say actually just along the way, this is also where we're going to be continuously bringing capabilities. So we don't plan for the announcements, we plan for it as fast as we can. We've got work that's coming in the next couple months, probably for Dreamforce, where we're going to be actually announcing support for other agentic partners and actually bringing just like we did bringing our APIs for Agentforce we're going to be bringing our APIs for other partners as well. So, yeah, stay tuned and listen in.

Speaker 2: 19:21

We'll have some great announcements both at Dreamforce and actually before that as well okay, thank you, andrew, for this nice conversation I really appreciate having you on our show I appreciate the opportunity and to all the listeners we look, I love, love to reach out and uh always ping me if you have any questions okay thank you, thank you oh.