# MuleSoft Keynote: Build a Foundation for Autonomous AI Auto-transcribed by https://aliceapp.ai on Wednesday, 18 Sep 2024. Synced media and text playback available on this page: https://aliceapp.ai/recordings/AtSYZ2LOJsUVzXh8fuDkmCgwn0xl6Miv * Words : 7,565 * Duration : 00:47:29 * Recorded on : Unknown date * Uploaded on : 2024-09-18 22:53:07 UTC * At : Unknown location * Using : Uploaded to aliceapp.ai ## Speakers: * Speaker A - 13.42% * Speaker B - 16.72% * Speaker C - 9.66% * Speaker D - 7.02% * Speaker E - 9.39% * Speaker F - 12.72% * Speaker G - 15.14% * Speaker H - 5.97% * Speaker I - 9.97% ---------------------------- Speaker A [00:00:00] Welcome EVP in GM automation and integration. Speaker B [00:00:04] Mulesoft Parham Kahlon. Speaker C [00:00:08] What an amazing story. What an amazing story. And such a great way to kick off the keynote. Good afternoon, everyone, and welcome to Dreamforce 2024. Now, uh, before we begin, I want to give a big thank you to our muleys, our flow, natics, our customers and our partners. Thank you all for being here today. Your feedback and insights have been vital to shaping our roadmap and product strategy. Today, we have a lot to share with you and an amazing lineup of speakers. First, we're going to return back to Dior story and learn more about their connected foundation. Then, to the lens of Ashton Martin, we'll take a look at how you can deploy agents and AI into the enterprise in an integrated fashion. And then lastly, we'll have Nvidia and our own Salesforce CIO, Juan Perez lead a discussion on how you can prepare for the future. And through all of this, we're going to show you our, uh, latest product innovation and demonstration. So first, to bring that video you just saw to life, I am so excited to welcome on stage Jean Charles Wardier, CIO of Dior. John Charles. Please come on up. Speaker D [00:02:02] Hello, param. Thank you for welcoming me, uh, here, uh, to Dreamforce. I'm very happy to be on stage with you to speak about your star and the technology azure. So let's talk about your star or app. Uh, let me tell you first why this app was born. It's very simple. We wanted to make our customer experience unique in our boutique, thanks to technology. So this app was designed for sales associates, not for our clients. To contribute to the magic of your experience. It's more than an innovation, it's a game changer for us. Uh, let me tell you how it works. Let's say, param, uh, that now you are Dior sales associates. Congrats. You are part of Dior team. Now, you have already the shoes I saw, so congrats. Okay. I'm a client and I work up to you. In store with Dior Star, you would be able to identify, uh, immediately my history. Don't worry, you have already given me your consent, so you would be able to see many information about me, my purchases, uh, my preferred products, my favorite stores, my hobbies. And as you were helping me, uh, in the store, you would be able to find any product in all its sizes or colors worldwide. Uh, you would be able to order from anywhere. Uh, I could pay instantly. And the product will arrive by magic. In one word, your star is your magic assistant. To offer me, the client, a customer experience even more personalized, seamless and smoother. Dior star helps us keep m mister Dior's promise to be the designer of dreams. Speaker B [00:03:56] Wow. Speaker C [00:03:56] Wow, that's amazing. John Charles, you and your team, you've delivered this amazing, like next generation of experience that's redefining how retail is done at deore. This is amazing and I experienced it firsthand this past weekend. So that was awesome experience. Really, really great. Now you've got this awesome digital experience for both employees and customers. What's next as you look to the future? You know, how are you going to amplify this even more? Speaker D [00:04:26] I think that, uh, we are just at the beginning of your expense. Powered by technology today, Dior Star is based on Salesforce core platform. We use Misoft also to share data, a lot of data. Ah. And dual star is a huge achievement. 90% of our retail transactions are managed by Dior Star. So it's huge, huge achievement. Uh, but with no doubt, the future relies on gen AI that offers many possibilities. Imagine that you can ask Dior Star, uh, what would be the best product and the unique experience for Jean Charles right now. Thanks to Genai and all the data Diorstar has about me, the answer could be, Jean Charles is the head of technology at Dioreghe. He loves sneakers. He is going to be on stage with Param at Dreamforce. He needs to represent the brand Dior brand, but he needs to represent the technology. So the best product for him is the B 27 black Dior sneaker. Speaker C [00:05:33] And they look pretty good too. Speaker E [00:05:35] Yeah. Speaker D [00:05:37] But for us, uh, Jenai will never replace the human connection as, uh, with the client. Uh, we always look for the perfect match between, uh, our craftsmanship, savvar fair and technology tradition and modernity. To make it short, our big challenge is, uh, uh, our big change is to really magnify the past because our customer wants the past, the past of the other history of, with technology. So that's the future. This is our goal. And your star is the first step to reach this goal. Thank you, param, for this great moment. Speaker C [00:06:18] Thank you, judge Rawls, we're so glad to have you here. Really, really appreciate you making the trip here. Thank you so much. Speaker B [00:06:25] Thank you. Speaker C [00:06:33] Now, as we look at the roadmap, we anchor ourselves on three persistent truths. First, data is the new lifeblood of every organization. Yet 72% of the apps and the data in the enterprise are disconnected. Second, we all recognize that AI is a game changer. Yet if you ask it, leaders in an organization. They'll tell you that 95% of the times m integration issues is what's holding them back from deploying AI more successfully in the enterprise. And third, we know that cybersecurity is the top, most reputational and financial threat to a company today. Did you know that a third of cyber attacks this year were targeting a company's APIs? And with APIs being the gateway to LLMs, this thread is even more pronounced now. Yesterday, Mark unveiled Agent force, our latest platform that will allow humans to work together with autonomous agents. With Agent force, work will never be the same and we can scale like never before. But, uh, you also know that at your organization, underneath the surface, there is a lot of complexity. And it's not just you. On average, an organization has 991 different applications. Each one of these applications has its own update schedule and own integration needs, their own security requirements. And you're probably wondering, how are you going to get all of these apps and systems and your processes? For Agent force to speak the language of your business, manually coding all of these applications and data into one is simply isn't viable or even sustainable. The industry best practice is to move towards an open and secure API architecture. That is why we believe that in the world of AI is truly the world of APIs. So I'm really excited in that spirit to announce and share with you our latest product roadmap. And we'll see much of this in demonstrations today. And uh, there's truly something for everyone here. If you're a developer, we're investing in the core Mulesoft platform and introducing new capabilities like event driven architecture and Async API. If you're a business admin or system administrator, we are introducing continuous innovation in Salesforce flow, but also low code integration with Mulesoft so you can build those integrations easily without writing code. And for everyone, we're investing in AI and agents. So today, to showcase all of this latest innovation, we worked with one of our customers, Ashton Martin, to build an amazing car buying experience through which we'll showcase most of our product demos today. So let's roll the video. Speaker B [00:09:54] Here at Aston Martin, the vehicles are. Speaker F [00:09:56] Quite fantastic in their own right, but experiences transcend that. Uh, we need to make sure that our customers experience is as perfect as our vehicles, but we had several disparate systems under the hood. It was a spaghetti jungle of APIs. Speaker G [00:10:12] Without the unification of the data on a single platform telling you how a. Speaker F [00:10:15] Customer is interacting with the brand, we. Speaker G [00:10:17] Can'T offer an ultra luxury experience. Speaker F [00:10:19] The team really need to overcome three main challenges, building out that unified platform reusability and speed. Mulesoft has allowed us to consolidate six legacy systems into data cloud to ensure trusted data quality across all of our customer touch points. For example, we log into customer configurator to spec out a vehicle, be that on the website, through a mobile app, customer interaction at a dealership. That single reusable API layer is integrating for each of those ecosystems, but it's also used in our own systems as well. All this information will be pre populated with your software. We have 100% API reusability to reduce our data entry points by 80% of what it was previously at six to seven times fast and smaller speed to market. As a developer, you love the integration, you love the technology, but at the same time it should almost be background integration. Landscape should just work and it just works. Speaker G [00:11:17] The beauty of a single platform is. Speaker F [00:11:19] To bring all of the capabilities and power of AI to all of our end users, enabling them to build a deeper, uh, relationship with the customer. With Agentforce utilizing data cloud, we can start to be able to answer, uh, more intelligent questions. You also. On top of that, we have that trusted element in a more seamless approach. Both agents and staff working in harmony, moving dealers away from a computer system and back in front of the customer. So they were part of that us and Martin experience to get humans back to doing what humans do best. Speaker A [00:11:57] Wasn't that amazing? I know that video made everything look so easy, but as we all know, building a purchasing experience like that is not easy. Aston Martin had to connect hundreds of applications across CRM, marketing, dealer management, supply management. The applications ranging from SaaS on um prem to custom, and with operations ranging in over 50 countries. They had to keep in mind regionality while thinking about scalability and reliability. So how did Aston Martin do this? Using mulesoft and AI, of course. So let's show you how. Artem, um, you ready? Speaker F [00:12:41] Woo hoo. Speaker A [00:12:42] M all right, let's go. First, let's meet Jerry. Jerry loves Aston Martins and he's almost ready to buy one. So he goes on the website and configures his dream car, a DB twelve coupon jet black. She's a beauty. And he submits the configuration. And this immediately kicks off a conversation with am. Um agent. Am agent is Aston Martin's personalized agent. Built on Agentforce, it can do a variety of things like answer questions. It could even schedule service appointments and test drives. So let's see how this conversation goes. Am agent says, hey Jerry, thank you for your interest in DB twelve, would you like a test drive with the configuration you selected? And Jerry says yes and it says great. How about this weekend on Saturday or Sunday, uh at your local dealership? And he says great, Saturday at ah, 03:00 p.m. and it says great, you're confirmed. M. You'll receive an email confirmation shortly. How seamless was that? Notice how am agent was not only able to answer questions but it was also able to take action based on Jerry's responses, orchestrating integration across many different systems. And I know how did they do that? So let's get into it. To do that we're going to go to anypoint code builder, our ide and this is where we're going to build topics. And topics are a combination of APIs and metadata that teach your agent how to take action based on your API. So we're going to first start by designing our API spec for test drive API. And notice something new here. Uh, now I can enable topics and actions and if I click that button it'll allow me to add a lot of details like the label, the description, the instruction, thereby teaching the agent how to use this API. And once I save this I need to implement the API now and I could do it from scratch. But why? Let's ask Einstein to do it. Hey Einstein, could you please implement this test drive API for me? And look at that. Einstein has now auto generated this implementation for me, saving me a ton of time. Thank you Einstein. And once I save this topic, this topic is now available immediately. Topic center, along with all the other topics that I've created. And now you may be wondering, how do I connect this topic with an agent? Great question. To do that, let's go to Agent Builder. And here in agent builder I can import all my topics that I've created, including my schedule test drive topic. Amazing. So by clicking on a few buttons now I've connected the topic with the agent, thereby empowering the agent to schedule test drives on behalf of the customer. Huh uh, but there's something missing. Remember when the agent knew to ask Jerry for a test drive and knew to do that because there's a confluence article that says if a buyer is interested but not yet ready to buy, a great next action is to encourage a test drive. That integration is powered by Mulesoft direct for data cloud. This salesforce embedded solution that is powered by Mulesoft helps you bring in unstructured data from knowledge repositories like confluence, SharePoint, Google Drive into Datacloud's vector database, which Agentforce has access to. So without any coding, I was able to bring in all these insights, all this knowledge into data cloud, thereby empowering my agent. Amazing. Oh, yeah. So what's up with Jerry? Where is he? For Jerry, it was love at first drive, and he just signed the papers to purchase his first Aston Martin. Congratulations, Jerry. So what would have normally been a long sales process has now been transformed thanks to mulesoft and AI. So I talked about a ton of innovation. So I want to do a quick recap. First is Mulesoft Topics center. This is the home to all our topics which teach our agent how to take action based on our APIs. And this will be available in Q one of next year. Next is Mulesoft direct for data cloud. And this helps you ingest unstructured data from knowledge repositories into data cloud. This will be available in the fall of this year. And last but not least is Einstein for anypoint code builder, which helped me implement my flow automatically. And this is available now. So all this innovation is wonderful, but nothing is more valuable than hearing directly from a customer. So to do that, I would love to welcome to this stage Matthew Randall from Aston Martin. Come on up, Matthew. Hey, Matthew. Thank you so much for being here. Speaker F [00:17:56] Thank you. Speaker A [00:17:57] I know you've been with, uh, Aston Martin for 16 years. Congratulations. Speaker G [00:18:01] Thank you. Speaker A [00:18:02] If you can remember, what was life like before Mulesoft and what were some of your challenges? Speaker F [00:18:08] So I think our story is very similar to what a lot of other people may have had. You know that I always boil it down to three core problems, situations try and solve. Right? And that already comes down to those three points. Two of them is very developer heavy. I come as a developer background, I know what those burdens were. M first was the amount of time it takes to actually develop. Coming from a.neT background, it's great, all the work you can do, but it takes a large period of time to try and get that next API, doing any form of extensions, building that up. The second angle is definitely still around development period itself. And that's onboarding new people is every single time you bring in a new developer or you bring in a new consultant is making sure that you have the documentation at hand to be able to train them, handhold them, explain where it is, because the ways of doing it is vast and all encompassing. Speaker A [00:19:09] Mhm. Speaker F [00:19:10] The third one is documentation, and this is API documentation. We need to make sure that anybody that is using any form of APIs, we implemented loads back when 2017 hit. We wanted to try and push this out. But all that documentation sat in. Swagger documentation.net application documentation, as well as across 15 other locations. New supply comes. And so where's my API documentation? Oh, it's over here, over here, over here. And that takes a fair amount of time to handhold even that process. Speaker A [00:19:44] Very common challenges we hear across the board from prospects. Um, and how has the experience changed using Mulesoft for your employees as well as for your customers? Speaker F [00:19:54] Um, I think the big thing is answering each of those questions in turn. Right. Development in of itself is a simpler landscape in the fact that it is a well established process. We know what we need to better do, which means that onboarding new developers as well as upskilling, we've seen some great stuff from earlier on on how quickly you can develop such, that low code pieces. We don't need to have all that reusable architecture and a copy paste of code sets, uh, that all exists. The other side of things as well is onboarding new developers. It's such a well established, we've got all of our muleys that was already being shouts out, they're there, uh, as well as all the partner network you can get and call those people on and upskill people really easily through various um, training methods. And that is what's really important. And thirdly is all of those activations that you get with exchange and the experience hub, right? You want to be able to have that discoverability. We have managed to break down a lot of that documentation period from a 2345 weeks experience, trying to lead a new supplier to utilizing them to a couple of days. Because here is everything nicely and discoverable. That's what we want. I think the big item that I absolutely love is success. As you saw from the video, you don't often talk about inspiration. It's the reason I added it to my job title. You want to be able to talk about it, you want to measure success. And I think that was really shown as part of our vanquished launch only a few weeks ago. Everybody always concerned that you got a bunch of APIs all coming in to have new leads, new opportunities, new orders, you don't want those, you don't want to drop new customers that may be coming in that have seen the new cardinal. From a great statistic that we had in 24 hours followed the launch, we would normally have something around 250,000 API connections, which is already quite a lot. We had over 1 million API connections during that uh, 24 hours period and we still had a success rate of 99.99%, which was all visible with insights. Speaker A [00:22:20] I love that story. Thank you so much, Matthew, for joining us today. We really appreciate it. Matthew, a round of applause. Speaker F [00:22:28] Thank you very much. Speaker A [00:22:33] Isn't that inspiring? Are you fired up? Um, and I know you're excited to hear about how Aston Martin is transforming other, uh, parts of their business, like customer service. So to do that, I would love to welcome to this stage my friend and my colleague, Vijay Pandyajan. Speaker G [00:22:53] Thank you, Ayan. Speaker B [00:22:54] Um, it's so great to be here. Now, raise your hand if you have a task or a business process that you absolutely can't wait to get automated. Yeah, there's a lot of manual stuff still out there. Now, did you know that it takes an employees 40% of their time still doing manual repetitive tasks that don't really add to the end customer experience? We can go a really long way to helping both our employees and our customers if we can automate access to external data. Now, integration has always been a developer for sport, but today I am so excited to announce Mulesoft's low code integration. It gives business users the power to access external systems without writing any code. And we've built it into the well known and proven interface of flow builder. Now, how about getting data from all those little bits of paper that's infiltrating and clogging up all your processes? Ever have that problem? Let me tell you about IDP, or intelligent document processing. With IDP, you can access data from documents such as quotes, invoices, and purchase orders. We can then use AI to extract, summarize, and classify this information, and most importantly, get it right back into the business process, where it can do some good. So, picking up on the Aston Martin story, let's see how Aston Martin could use these products to create a great, streamlined service experience for their customers. Let's jump back into the demo. Aydham, you ready? All right, so Jerry's been really enjoying his car. We're going to fast forward a whole year. He's racked up more than 10,000 miles on it. He's also got his eyes on these really cool neon blue wheel rims that he would love to have installed. Not that the Aston Martin isn't great. He wants to make it even better. And so he opens up his Aston Martin mobile app, and he puts in a request. Can I bring my car in for a service? Can I get these things plugged in? And once he does that, he's super excited, and we can schedule that service. Now, once the service is scheduled automatically, we've got a business process being kicked off in flow orchestration. What this lets you do is to capture all of this information and walk you through all of the different steps in the process. Mulesoft now supports event driven APIs based on the Async API standard, which facilitates real time sync of vehicle telemetry data right from Jerry's car into data cloud. Now, orchestration can access this telemetry data in data cloud, and we can tell exactly what service, uh, needs to happen on Jerry's car. Isn't that cool? And then we can also classify how that repair is going to get covered. Is it going to be covered by warranty? Is it going to be covered by insurance? Or does the customer have to make a payment? For example, we can tell from the telemetry data that there's a broken proximity sensor on the front bumper. This is going to be covered by insurance. So let's go see how we're going to process this insurance claim. To do this, we're going to use an invocable action to reuse an asset, an API that was already built by the development team. In this flow, you're going to see, we're going to call a mulesoft API that can access the, um, integration API of an insurance carrier to retrieve a work authorization document. All of this even before the customer even walked into the shop. Now we're going to pass that document over into IDP, or intelligent document processing. And here you can see a business user can very easily capture information from this. We're going to map certain key fields that we want into variables so we can use it later. And we can also use Einstein to ask this document questions such as what was the amount authorized? And we can tell here with a great deal of confidence that that's the number. And we're going to use this number as we go through this process. Now, the next step here sometimes can be very tricky. We're going to try to access Aston Martin's order management and inventory systems. Anyone done that? Just like for fun. Now, this is where you would really need custom resources and you'd have to go through an approval process and wait weeks maybe to get approval. But now with Mulesoft's low code integration, a business user can build this themselves. And that's pretty cool. What you see here is a custom netsuite connector. It's already built for you. It's pre built. Uh, once you open up that Netsuite connector, we can check for inventory in those systems. We can reserve the inventory that we need, and you can see here that the variables that we captured earlier from IDP can now just be passed in. So it's kind of cool. We got that stuff out of paper. We're going to use this, we're going to drop it right in here, and we can have some interesting data mappings and other things that we need. There's a bunch of data that needs to come over. We don't want you to have to type all of that. We're going to just map that over. Plus, how do you get the data into exactly the format that you want? So we've got some powerful data transformations as well that can help you bring this over. Now, remember those custom wheel drums that he wanted? Our service team has already gone out, built out an estimate for that. In the next part of the demo, we'll crack open how the service team did that part of the work. But we have all the information we need right now to get back to Jerry and put him in the middle of our process again. We're going to let Jerry know that, uh, everything is available to him. And Jerry's pretty excited because he just gets pinged on his phone. It pops up in an agentic experience. Jerry's really thrilled. He wants this thing to proceed. He says, yes, please go ahead, get this service scheduled. And this brings us now to the last step, uh, of our orchestration. Remember, orchestration has been keeping up with what everyone's doing. And what it decides now is that it's got to send out a car. So Aston Martin does this really cool thing. They send out a local transporter to Jerry's house to pick up his car. Now, that's a true luxury white glove experience from beginning to end. Now, uh, the important thing I want to bring you back to is how data access was so important to unlocking this really cool experience and how we use that data access all the way from low code through pro code. We were using things, we were building things in there. We were getting a lot of people involved, business users, it teams. So this really becomes a team sport. Creating these experiences is a team sport. Now, uh, what we haven't talked about yet is how you can bring the same agentic experience to any application, any bespoke application that you might be building, and to take you through that part of the story about how to build an agentic experience completely, securely. Please welcome to the stage my friend and colleague, Simon Goddard. Speaker G [00:30:12] Thank you, Vijay. Good afternoon, Dreamforce. Now, for the final part of this demo, we're Muley's. Let's take on something even more challenging. Let's look at how we can secure and manage AI agents in a complex legacy technology environment. And as para mentioned before, AI security is paramount, especially in a legacy environment. And in this next demo, I'm going to talk about two products that we haven't seen today, and we're going to talk about how that can significantly reduce the risk of vulnerability. So first up, say hello to one of our latest innovations. Meet API insights. Treat this as your API command center. You can track API usage over time and even see the governance status of all your cataloged APIs. The second product we're going to show is Mulesoft robotic process automation. Here we can program robots to access and perform tasks on behalf of humans in any system, including legacy. So let's take a look at how Aston Martin can use these tools to reimagine their supply chain experience with AI agents. And you might remember our good friend Jerry ordered some neon custom blue rims. Pretty snazzy. And it was allocated to a service specialist, Hannah. Now, Hannah operates in our supply chain and it's a pretty manual task, not very strategic for Hannah, so she would love some help from an AI agent to make this easier. Here's the challenge, though. In the supply chain we have disparate technology systems, different types of integration. For example, our first supplier uses a legacy mainframe system that doesn't even talk to APIs. Shock horror. I know, right? We then have our second supplier who uses a heavily customized version of SAP. Let's show you how any integration into any system is possible when you use a unified platform from Mulesoft. So let's tackle that legacy mainframe, shall we? We are going to leverage Mulesoft RPA and we are going to program a robot to open that mainframe to search for inventory, negotiate pricing based on logic that you set, and even place the order. If Hannah wants to go ahead. That's pretty cool. How about that second supplier though? Here with SAP, we need a custom API integration. So we're going to go back to Einstein for anypoint code builder and we actually use Einstein to help with the integration. They are going to leverage our, uh, out of the box connector for SAP, which significantly reduces development time. And then we have a flow in place. And, uh, because of the development time saved, our developer actually adds in a slack notification flow here. So Hana can see the order status in her flow of work. This is a game changer for Hana. Now imagine replicating this process across all of Hana's suppliers. We can then surface those integrations and APIs into agent force, just like Arjun. Ah, demonstrated earlier. And now our, uh, AI agent has all the fundamental building blocks in place to navigate the supply chain. That's pretty cool. But we didn't talk about security, right? For example, Aston Martin doesn't just want anybody like me using this agent to order fancy neon blue rims. They want authorized users like Hannah to access this. This is where Mulesoft's API management capabilities come into play. And we can leverage API insights to see all the APIs that we have cataloged. And if we scroll down, we can click insecure APIs. And here we can see all the APIs that have no policies applied are what I like to call the naughty list. There's one in every organization. Okay, let's filter this further. Let's have a look at the AI specific APIs using our filtering. Okay, let's do something about this auradum, huh? Let's apply a policy to a production API. All from this dashboard, we click Add policy. And here, there are dozens of policies that will be available soon in your mulesoft policy development kit. But in this instance, we have a, uh, naughty list of APIs. And I have a nice list of users like Hannah, who I want to authorize. So I'm going to click client id enforcement. I click next and I can then apply the policy with clicks, not code, straight into production. That's pretty game changing. So let's have a look how this feels for Hannah. She simply talks to the internal am service agent. She asks them to order these pretty snazzy blue rims. And just like that, the agent invokes the APIs and the integrations we've just created and orders the part. When the part arrives, our, ah, service technician uses the same internal service agent to say to Jerry, it's time to pick up your Aston Martin. There we go. Let's see how this feels for Jerry. For him on his phone, it feels seamless. In the same Aston Martin app, he simply gets a notification and let's see how he responds. He wants to drop his vehicle. Oh, my God. Are you seeing this? Jerry wants to give me his Aston Martin. And, um, well, here it is right now. Oh, my gosh. That is amazing. Thank you, Jerry. All right. You know, as my mum says, it's the thought that counts, right? So, uh, these are fun sized. And if anyone knows Mulesoft, you know we love a building block. Uh, would anybody else like a fun sized Aston Martin? I could see some. You know what? This is my Oprah Winfrey moment. Take a look under your seats. We have planted some golden tickets in the crowd and if you are the recipient of a golden ticket, wave it in the air. You can, uh, head to the mulesoft lodge downstairs to pick this up. Can I see, I can see some golden tickets here. Speaker F [00:36:29] Yeah. Speaker G [00:36:29] Round of applause for our winners. Yes. Congratulations. And there are more of these surprises to be won in your mulesoft sessions. Vijay, can you look after that for me? Speaker H [00:36:43] Huh? Speaker G [00:36:44] I'm watching you. I'm watching you. We love to surprise and delight. Okay, that was a quick spin through some of the latest innovations from Mulesofthe. We didn't have time to cover it all, though. As Param mentioned, there are heaps of innovation coming to you to help you revolutionize your employee and customer experiences through integration, automation and API management. Now, to bring this to life, let's hear from some leaders who are driving the AI strategy at their companies. To facilitate this conversation, please welcome to the stage Salesforce CIO Juan Perez and our special guests from Nvidia, Ashwin Jhar and Ramar Akaruju. Thank you. Speaker B [00:37:35] Hello. Speaker E [00:37:36] Good to see you both. Hello, everyone. It's great to be here with you. Thank you for being here. In this session, I first have to start with a quick public announcement. I want you to know this. I'm, uh, the CIO for Salesforce and I'm one of Mulesoft's largest customers. Um, I will tell you that much. We run our APIs, we manage our integrations, we manage our RPA, we manage a lot of our document processing in the enterprise using Mulesoft. And I'm a huge fan, I want you to know that. But you don't have to believe me. The public announcement is that when you get an opportunity, you go to the floor at the campground, you go to the Salesforce and Salesforce booth, and you'll see some of my amazing team members who truly believe in the power of Mulesoft. And you can see how we actually power this company using Mulesoft. So I invite you to do that. I think you'll learn quite a few things from them and you can certainly ask all kinds of questions about our journey using Mulesoft. Now, in terms of journeys, thank you both for being here. You're just amazing partners. Thank you for what you do. You have some incredible titles in your jobs. Vp of it or AI for it. I think that's phenomenal. And then you have also a really cool title, managing enterprise productivity engineering in the enterprise is just phenomenal. But now you've been in this journey of implementing AI for quite some time, one of the things that I think most people here would want to know, based on the lessons that you have learned throughout this journey, what are some of the things that we all need to adjust to in terms of new operating models to manage in this new world of AI? Either one of you can start. Speaker I [00:39:11] Thank you, Juan, for having us here. Wonderful keynote panel going on. Um, so, yeah, we've been on at Nvidia, we've been on our journey to deploy AI into pretty much every business process within our enterprise. Uh, take it, supply chain, sales, marketing, finance and so on. And there have been, in the past two years since chatgpt came out, which put everything on steroids, a lot of learnings. Um, if I have to summarize, I would say, you know, Ashwin and I work very closely together. We came up with this, a simple mnemonic to remember all our learnings facts. It stands for f for freshness. And I'll talk about how to bring freshness in the enterprise, bring the fresh content to the enterprise. A for architectures and agents, and c for costs. How do you manage costs with your LLMs and generative AI based solutions and AI solutions in general, and t for testing what it takes to test generative AI solutions and s for security. So freshness just kind of quickly mention, we all have been through this journey, just an LLM won't do to bring fresh content in the enterprise, and we bring enterprise perspective. And so you bring retrieval, augmented generation, and there are a lot of things that one has to get right, as you do. Do it yourself. Uh, you are asking in the keynote panel not to do that. And we did that. Some of Nvidia technology and a lot of learnings there. But retrieval, augmented generation for bringing fresh content, and for architectures, of course, have flexible architectures that are configurable, modular, because everything is changing so often in the enterprise. And agents, uh, with agentic architectures that enable this modularization and connectivity with APIs. And, uh, as you do the LLMs, make sure, as you leverage LLMs in the enterprise, make sure that the costs are, um, managed, because, uh, the costs could go up significantly into millions of dollars in the enterprise as you deploy a lot of generative AI solutions. So how do you use smaller model versus bigger models and when and what to use? So there's a lot of learnings there for us, account for enough amount of time for testing and finally, security. A lot of learning's there because enterprise content cannot do without proper access control and governance so in short, facts, easy to remember. Speaker E [00:41:27] I love it. That's a great way to remember, right? How to manage and implement this great technology and talking about facts. Ashwin, you're in the world of productivity. I love the word productivity, actually. And, uh, you know, it's one of those things that as professionals, we have the obligation to continue to drive and push for. When you look at the journey that you've been in terms of implementing AI in the enterprise and you think about productivity, how are you measuring the value that is coming from this technology? Speaker H [00:41:55] Yeah, and that's a good question. Um, you know, we have been like, we always think about an AI, and I think yesterday's keynote, someone also mentioned that is this system breathing system, which will one day come alive and solve all the world's problem. Um, actually, it's the contrary to that. We have been targeting all the use cases that we could see and start experimenting quickly. Understand, pivot from there, and you will see that slowly and steadily when you keep solving this automation challenge. And now AI is going from just insight to actions like playing from the agent force perspective. All of these automations can talk together and bring, finally, the digital intelligence home. And when you keep solving, like each of the use cases, I'll start with. We started answering our customer service help desk questions directly through the bots. I mean, it solves one problem where you can be more productive while working on something important. And your bot is answering questions and doing the augmentation with the human in a perfect way, because you also want to make sure that this product evolved along with the feedback from the customers, from the humans. And this collaboration of Human AI is at the center of it. But how we think about it is that in every use case by use case, start implementing this, start seeing the results. And for example, for our code critic, um, where we can review codes while somebody is sleeping, you come to the office and your code is reviewed, and with all the comments done, so you don't have to wait for those things. And it can really add up very quickly for any organization if you talk about the value. Speaker E [00:43:25] I agree. And, uh, early, uh, days, I was hesitant to actually speak about the improvements in coder efficiency that we have seen. I was kind of skeptical as to whether this technology would really materialize in productivity improvements. But just in the last period, in my own shop, we promoted over 60,000 lines of code to production. 26% of those lines were generated by our cogenerator, which is a significant improvement from where we were just six, eight months ago. So this is going to continue to materialize into real benefit for the company. And Rama, the question that I have for you now is, as we think about deploying more of this technology, we obviously have to have the right architecture in place. We cannot live without great architecture to support this phenomenal technology that we are talking about here at Salesforce, and obviously the value that comes from it. How are you approaching the architecture problem? Speaker I [00:44:18] Yeah, uh, it's a question that's near, uh, and dear to my heart. You know, there used to be a time when we used to think that you go on vacation for a week and you say, oh my, too much is happening in the world, but now you go for lunch, new model has come out, and whatever you have in your production is outdated already, right? It's changing that fast. So when the world around you is changing that fast, how do you make sure that your architectures are dynamic and adaptable and flexible? And that has been one of the key tenets as we were building many of the things, uh, at Nvidia, we wanted to make sure that we built on a platform where we can switch LLMs whenever we needed to. We can adapt prompts like the prompt builder that you all have been showing with agent force and Einstein platform, and that we can even change the retrieval model, information retrieval systems, you know, you can, there are many out there with vector database searches, so you should be able to switch, swap those things out. So pretty much, you know, in a running car, you should be able to take out the chassis, you should be able to swap out the engine, you should be able to swap out the tires and everything. That's how your architecture has to be. And, uh, you know, all the core tenets of fundamental software engineering, good principles of modularity, of, um, better configurability, of proper governance, all of those things, they don't go away. They are still very much pertinent as we build AI platforms. And so we just went back to those core principles and made sure that as we were building our solutions, that we built on a platform that's replicable and scalable, that also allows us to take this platform, give to others in the organization so that they can build their applications faster, so thereby democratizing AI, which is what you all were talking about as well in the keynote and with your platforms. Speaker B [00:46:01] I love it. Speaker E [00:46:02] Great, thank you. Last question, quick answer. Now we got to get ready for this new world. And as driver of productivity in the enterprise, how do you prepare for this new journey that we're into now? Speaker H [00:46:13] I would say that the most important step of getting started is getting started. Um, it's so important. And once you get started, you will start realizing what are the solutions, use cases and how these whole things come together to deliver that end case solutions which you're targeting to do. But it's very important to put your hand in the sand, get experimenting faster. Like Rama mentioned, create those agile architectures so that you can move faster. I'm guaranteeing you tomorrow there will be another amazing thing which will come out, which you can utilize out of the box. So either you're buying or building, just get in and you will find use case by use case. All this excitement is we are living in an exciting world. We are looking forward to what can be possible which was not possible yesterday. Today we can do it with a click. So it's going to be more exciting. Speaker E [00:47:00] I love it. Well, lots of lessons learned, right? In just a few minutes, facts, matter. Measure, measure, measure. Because the day will come in which we will all be accountable for all these AI stuff that we're working on. Obviously, we got to get our teams ready for this new world that we're going to be living in. And then at the end of the day, we got to bring it all together so that we can create true value for our enterprises. Thank you for your partnership and doing such a great job today. Appreciate it. Thank you both. Take care.