Build the Agentic Enterprise | Developer Keynote, Dreamforce 2026

Please welcome SVP Trailblazer Relations, Salesforce, Christoph Konrads. All right. All right. Good morning, everyone, and welcome to the Developer Keynote. To all of you in the room and everyone online joining from all over the world, welcome. Now, today is about what's next for developers. But before we start, I want to take a moment to thank you. We have MVPs in the room. I see some golden hoodies, community leaders. To all of you, thank you for everything you build and everything you do for this community because you really make this community unlike any other. And this year, we are celebrating 20 years of Trailblazer Community. 20 years. That is incredible. Yes, but what makes this community really unique is you and what you do. Without even being asked, you bring people together at community conferences and user groups. You lift each other up, you answer questions, you mentor, and of course, you shape the platform by giving us the feedback that we really need to hear. And all that energy translates into real impact. Because you build what the world runs on. Now, seriously, all of you, the companies you work for all over the world, big and small, they run their business on Salesforce. So every single number that you see on the screen here is something that you built. So you can give yourself a big round of applause for that. Now, Speaking about what you build, I don't know if you noticed, but how you build is changing really, really fast. Just last year, just last year, you were using AI for things like, write this class or explain that piece of code to me. Really, really small tasks. It took seconds. It was coding assistance. But today we are asking AI much bigger things, Maybe something like build an entire system that delivers this outcome. And then we attach a spec. And then a coding agent starts planning and writing code and testing. It can even open a pull request. And then you can set up another agent to review that pull request. And it can go on and on like that for hours, iterating toward the outcome. Now, how many of you have done that? Give a coding agent a goal and let it work for hours, right? That is the shift from assistance to delegation. Now, don't worry if you haven't done it. We're going to show you how today. But all of this is raising an obvious question that I know many of us have been thinking about. If AI writes the code, what happens to the developer?

Right?

Tension. And the answer is not a smaller role. AI is actually expanding the role of the developer because you don't just give a coding agent a goal and let it run wild for hours. No, you direct AI. Yes, you set the goal, but you steer AI toward the outcome. You architect, you define the coding patterns and the best practices that you want to use. You connect everything, the context, the capabilities, the enterprise data, and of course, you govern. You set the guardrails, the controls so that we can deliver trusted AI. So as AI is taking on more of the execution, we own more of the outcome. And that gives you more leverage, more scope, and more impact. But you know what? That is only half the story. Because every technology shift raises the bar, raises the aspirations for what software can be. Take the web, for example, the internet. All of a sudden, we had a brand new category of applications, customer-facing applications, and every company in the world wanted to connect directly with their customers on the internet. Now, PC, cloud, mobile, same thing. They all unlocked entirely new categories of software. And every time that happens, the role of the developer is getting bigger. That's the other half of the story. And AI is the next shift. And once again, it raises the bar and it's changing what we can build. Because for years, in fact, since the beginning of software, we build applications with predefined capabilities, predefined logic, and predefined UI. But agentic software is different. It can understand, it can reason, and it can act. And once again, that's unlocking entirely new opportunities. Now, when software can reason and act, it doesn't have to live in traditional applications anymore, right? So AI is also changing where software shows up. And you heard it in Marc's keynote on Tuesday. Right? With AI for us, users can experience software in entirely new places: cloud, Slack, coworker, everywhere. Now, here is the opportunity for all of us. As software moves to these new interfaces, the custom solutions that we build can and should move there too. That's the opportunity. That's where it's going. So how, what, and where, it's all changing really dramatically. And to build all this, you need more than a collection of AI tools. To build this, you need a platform with trust and governance, security, enterprise data and context, deterministic logic and actions, AI, all this working together, integrated, and able to show up anywhere. That is the Salesforce platform. And today, we're going to show you how to use it to build the agentic enterprise. We're going to show you how, what, and where. And to get us started and tell us how to build with these new agentic workflows, please welcome Principal Developer Advocate Mohit Srivastava.

Mohit.

Thank you. Thank you. All right, thank you, Christophe, and hello, everyone. How we build with AI is changing really fast, but today, I'm excited to share with you the development workflow that I use every day now. We'll start off with coding agents, and I'll show you the Salesforce Development Plugin for Claude. And next, I'll show you how to work with Lifecycle Center. And we'll do all of this through a fictitious company, Radiant Homes. Now, Radiant Homes builds smart home devices, devices like security cameras, smart lights, and so on. Last week, they deployed a firmware update to a lot of devices. And as builders, we can all empathize, right? Despite all the things, things can go wrong, right? So the firmware update broke a lot of devices, and a lot of houses went dark. So our challenge for today is to build a failure detection system for Radiant Homes. But before that, are you all ready? All right, give it up for Renee and Sean, who are going to help me drive the demos. Come on. All right, so I am in Salesforce project, and I have Claude Code installed. Now, I know a lot of you work with different coding agents. And thanks to our headless toolkit, you can bring any of your coding agents. We support all of them. But if you're a Claude Code user like me, I'm excited to share that now you can install Salesforce plugin for Claude. It's generally available, and you can get it started right now. Now, with Salesforce development plugin for Claude, your coding agent gets secure access to Salesforce development. And it has all of the lifecycle tools built into it. And it comes prebuilt with all of the skills. You can see there are a lot of skills, and this teaches our coding agent how to build apps and agents on Salesforce platform. All right, but remember, we are building an end-to-end feature. We are no more building those small Apex or LWCs, right? We are building an end-to-end feature. And to build an end-to-end feature like this, I prefer to create a specification file. And to create a specification file, I wrote my own skill, write spec. So if you're repeating any prompts or any workflows, it's a good idea to package as a skill. That's one of our job roles today, to write and maintain the skills. Now, if you look at this write spec skill, it has tools. We give platform metadata grounding MCP to this write spec skill. With this tool, a coding agent can get into the Salesforce org, and it can access all of the metadata as context to create the specification. All right, so now that we have this write spec skill in action, let's put it to work. Back in my Claude code, let's write a prompt here. So here I'm asking the coding agent to create an implementation specification for our feature, and our features are proactive firmware failure detection, and we are asking the agent to build a visual control console. And at this point, you provide all of the technical knowledge that you have as a prompt and all of the business domain knowledge that you have gained over the years. So, and you can see we have the write spec skill. So let's go ahead and execute this prompt. Now, this is where all of the magic happens, because at this point, the coding agent is going to our org, is getting all that metadata, all that processes that you have built, all of the events, all of the schema that you have configured. And once it has all of this, thing, it's gonna create a specification for the implementation. But this is gonna take some time. This is gonna take some time. So just like one of our cooking shows, as we told you, it's gonna take hours sometimes. So just like one of those cooking shows, we'll fast forward it. And here is that implementation spec that the agent created. And you can see here that the agent first spelled out all the requirements. So we can align with the agent here. And you see that the agent identified that in our org, there is already a Data360 component, Calculated Insights, that's calculating the fault failure rate for the firmware build. It identified there are platform objects that it can reuse, and there is already a trigger actions framework in the system. So instead of reinventing from scratch, the agent is smart. It decides, yes, I'm going to reuse all of it, and proposes an architecture. And here's the architecture. And you can see here that the agent is recommending that we use platform event to send the signal from Data360 back into Salesforce for those Slack alert notifications. And then you can see it also proposes that all of the business logic, Here, you can see it's recommending that we create all these fields, all these security rules, and all of the business processes. And at this point, our job as developers is to use our thinking hat, all of our experience and practical skills that we have acquired over the years, and use our human judgment to go and verify this and iterate upon it. Might take a couple of iterations for us, But once we are happy, we can take this implementation spec and do our implementation. So let's go back to the Claude code. All right, here, let's bring the prompt to the implementation— for the implementation. Now, I want you to pay attention to this prompt. There are 2 important things here. So first, I'm asking the coding agent here to spawn different subagents. And each of the subagents can use skills from that Salesforce development plugin. I'm also giving the coding agent the tools, like same tools that we use, Code Analyzer, Apex Guru, Apex Test, Jest Test. And also, I'm giving the coding agent a goal. In this case, it's 90% test coverage. So the agent can get into the loop, it can keep working for hours till it achieves that objective. So let's go ahead and execute this prompt. Now, I don't want you to be sitting here all the keynote and just watching through the traces, agent traces. It's gonna go and work for hours, right? So again, just like one of those cooking shows, we'll have to fast forward it. And here's that initial pull request that the agent created. You can see it built all of the Apex classes, custom metadata, flows, objects, All of it. But the challenge is, how do we review all this work? How do we review this big pull request, right? We can use our human judgment, of course, our skills that we have acquired, and compare with the specs that we generated, but we can do better. We can throw agentic capacity into it. Here we have configured adversarial agent. The job of this adversarial agent is to challenge our implementation. And here, you can see it's flagging. It's flagging that there is an architectural risk here. It's flagging that this architecture is not scalable. There are things that we need to fix. And it not only advises us, if we scroll down, it also shows us the code changes that needs to happen for this. And here, we can use that Apply Suggestion button and right away apply. And this is all powered by the skills in Salesforce Development Plugin. All right, so now that we have, we know that this is the issue, we can go back and work on it. We went back to the agent last night. We're still working, right? Developers, right? We love to work sometimes. And me, Renee, and Sean, we all worked together, used our own expertise and real skills. And you can see that we fixed all of those issues that the agent helped us through, of course, but— That's awesome. Also, sometimes we steered the agent in the right direction. And here you can see we have fixed those. And if you go to all of our test cases here, you can see all of our code analyzer tests pass, Apex tests pass, LWC Jest tests are passing, and we have some trust to merge this. And trust did not come just from the coding agent. It came from the tools that we gave to the coding agent to self-verify. It came from the spec we wrote. It came from the human judgment and the practical skills that we have all acquired over the years. Now, this looked good. We merged it. Let's show this in action. All right, live demos, right? Hopefully, the demo gods are with me today. So this is a platform event to simulate the failure signals. Let's go ahead and execute this. We should see, there it is, the Slack notification confirming that now our reps will have this notification whenever A failure happens. Remember, we also asked the agent to build the visual console and all of the Lightning components. And there it is, that entire app built by an agent. It built all of the Lightning components that are required. All right, so this was one big feature. We guided the agent. I showed you the developer workflow. What should we do when the agent is doing this long 1-hour task? Have you thought of this, right? Get coffee, yeah, love it. While I'm sipping the coffee, right, we know Radiant has a lot of backlogs. They have a lot of small work items, just like all of our projects, right? There's documentation to write, there are bugs to resolve that come in real time. There are all these small bounded tasks. But how do we scale ourselves? How do we increase our leverage to work on these items? Can't be terminals. I can't have hundreds of terminals and just swiping through it, not the right developer experience, right? I'm super excited to share with you our brand new offering, Lifecycle Center. This is our Lifecycle Center, and there's the team that built it. All right. So within the Lifecycle Center, it comes with all of the predefined agents, agents to do architecture, agents to do pull requests, agents to help us with build, and soon they're going to add the custom agents. So you'll be able to create your own agents. Here, I can create orchestrations for bounded tasks like bounded small work items. So let's go ahead and create one. So when we triage a work item and make it ready for agent to build, We want our builder agent to pick it up. And here we can provide some prompts. And let's save this. All right, now we have this orchestration configured. I'm excited to show you how to sort of put it in action. Let's try it. So there is a real bug that has just arrived, that 001 bug. Let's drag it and make it ready for agent to build. You'll see right there, look there, it's building. There it is. Our agent is building in real time. It's using the tools and the skills that we provide to the agent. It's cloning our repo, it's creating a new work item tree, and there it is, it opened a pull request. There is our pull request and our reviewer agent is already reviewing this work. We went from a work item to a pull request all autonomous, All autonomous, but still supervised and controlled. How about that? But we know— hang on a bit, hang on a bit. We know agents drift. Agents also drift. They drift. And who fixes it? Developers, right? But check this out. Check this out. So here we are in Workbench, and every work item that the agent works through, we can click through and see the agent's execution logs. So here, if the agent drifted, we can do further conversations, and we can change models. Yes, there it is. There is a model picker. So if you need more control, more flexibility, you can scale up the models. How is it? What do you all think? All right, now we also know that we need to measure the efficiencies of these agents. It's not about just getting them to work. We will optimize it, we'll make it better, but we need to measure. So Lifecycle Center comes with analytics. And here you can see important metrics like lines of code accepted. You can see the hours saved. And this is definitely going to help us measure our agents, make them better. How about that? How about analytics for all of you, especially CIOs out there?

All right.

So that is how you build with agentic developer workflows. And with that, back to you, Christophe.

All right. Thank you so much, Mo. That's incredible. That's really how we build now. We guide AI, but we own the outcome. But you know what? How we build is not the only thing that's changing. What we build is changing as well. And to tell us how to build agentic systems, Please welcome Senior Director, Technical Content Engineering, Stephen Chandler-Garcia.

Thank you very much, Christophe, and thank you, Mo, for taking us through that. I mean, a lot of us have been using coding agents every single day, some of us even overnight. And I want to show you that you can also take that same exact capability that Mo showed us and use it in any agent. And that's why we want to help you get started by bringing back the free tier of Agentforce Vibes We heard your feedback, and we want to make sure that everyone can get started as quickly as possible. Now, we were talking about coding agents, but I want to talk about how we can take that same idea and bring it into the apps that you build, okay, into every workflow. Because in the same benefit that you get yourself as a developer, we can take that to our users, to our customers, to everyone who's using our apps. And we're going to do that using AgentScript, Agentforce Testing Center, and Agent Analytics. Now, we're going to take and continue that story. Mo was talking about how we could manage outages with a trusted agent, and now I want to take it and bring it into that application, and we're going to do that in a demo. And we're going to start right here where we left off. Okay, here we are in that Radiant firmware and product application. We have Agent Force right here inside of the app. We have their home operations agent that has access to all of the same exact data that the users do across the org. Now, this is us calling that agent from the sidebar, but you can access that agent from anywhere inside of the app. For example, we've been able to invoke that agent here from a Lightning web component so that we can take the same processing capability, that same insights that the agent provides, we can bring together the data and show it right here on the screen in the app. Now, this is just an agent that's in Agent Force, okay? We built it in the Agent Force builder. You can see here it manages firmware and outages, energy and demand, homes and devices— about everything that you think it would. However, this agent is backed by AgentScript, and it's crucial to understand that everything that sits behind Agent Force is AgentScript now, okay? It allows you to use natural language to write the scripting language but in a way that a machine understands, a way that can provide context and guardrails. And for example, this agent can do, like I said, all of those things. As you saw inside of the builder, we have all of these instructions here that tell it exactly when it should access a capability, for example, homes and devices or firmware updates, and it's able to tightly guardrail when it accesses one of those capabilities. Now, this agent breaks down into multiple discrete subagents. So here we have our Home and Devices SUM agent so that we can search for devices, we can search for homes, and we can see that we provide those strict instructions and conditionals based on when to access this. Now, we need to make sure that when a customer— when an outage happens, okay, we're handling it with very high levels of sensitivity. And so what we want to identify is whether or not if a customer has been impacted by an outage, whether they're due a refund. And we can use that and the guardrails here To make sure that only if a customer meets that threshold, we're able to do that. Now, we want to extend this agent, and we're going to do it with Agentforce Vibes because obviously it can also build agents. And we're going to do that by extending it with additional context. If we want this agent to function correctly, we need to bring in the data from Data360 into our agent so that we can then identify real-time devices and faults as we go. And so we're going to go ahead and run this prompt. Once again, this could take a couple of minutes, so we ran it earlier. And you can see here, we actually updated the agent here with that data from Data360. So it's able to identify actions in the org and query them in real time on that real-time layer to get it into our agent itself. Now, we can also use Agent Force V to test our agent. Once again, we said there are sensitive cases here. The guardrails are crucial. And so we can ask the agent to test itself. Okay, we're going to ask it to just run through a series of tests to ensure that the agent can perform and get the expected outcome no matter the utterance that you're putting into it. And we're going to go ahead and run that. But to give you a better idea of what that looks like, we're going to go ahead and run the Agent Force DX plugin here to just run a single simulation. And what we want to do is just ask it a basic question about a device. So let's go ahead and check out this outage. It's going to instantiate the agent, Okay, and it's going to send this prompt through that the agent can come back with a response. And you can see here it's been able to go and query the record, okay, find out the status of it. But more importantly, it's going to show us all of the thinking that the agent did, okay, every variable assignment, okay, every bit of reasoning. We're able to take that and break it down into data so that we can analyze exactly where we can improve our agents. This is the exact kind of data that will come back when you're doing this at scale to manage and track those improvements. Now, all of this data ends up feeding into Data360, where we're able to categorize and log all of the agent interactions so that you can track and manage this long-term to ensure that your agents are performing properly. You can see all the way down to the level of tracking which subagents are used, which actions are used, and then get into each, each individual turn. This same exact data can be brought into Agent Force 5's. Now, you can see here it's ran our single test. So we ran it across 10 different iterations. We have our test report that we've created here. We just asked Agent Force 5 to create us a nice HTML-formatted report so that we can visually impact and inspect what's happening. And we can see how we can then improve the agent. But like I said, we can even do this with all of the data that exists inside of Data360. So we can say, hey, can you pull the recent history, the recent session traces, And then let us know how we can improve this. And Agentforce Vibes is able to go in and pull that data here so that we can see where it can find the data, go and pull it from the environment, and then give us feedback on how we can improve our agent's performance and see that scale over time. So now we know that if we have a very sensitive issue, that the right guardrails will be held and the customers will be impacted correctly. So once our agent's ready to go, We're actually going to go ahead and talk about exposing it in additional surfaces. So we've taken our agent. We saw it inside of Agentforce. We saw it inside of a component. But I actually want to expose it to any agent. We want to extend our Agentforce agent into something like Claude. And we're going to do that using the Salesforce hosted MCP servers. Now, the hosted MCP servers allow you to add tools from all over Salesforce, whether they be Agentforce agents, Apex actions, flows, prompt templates. All of this is available to expose, to give to agents via MCP servers. In this case, we're actually going to give our Agentforce agent that capability. And so we have our Radiant HomeOps agent here. We have a single invoke method that we want to add as a tool to give access to additional agents. We're going to go ahead and save that, and we're going to activate it. Now, we did make the connection earlier so that we could demo this here. And there we go. We have our agent ready to go. Let's check if those guardrails are gonna hold. So number one, let's pull up an account and let's see if we can get all of their account details and all of those real-time device details from Data360. And you can see here, initially it searched for the available tools and then it called that invoke method, and it's been able to go and pull that customer as well as all of the devices that live in their account that are being pulled from Data360. We can see they have a device that's offline, a couple of issues here, and we have a nice summary of this. But where this really comes in handy is that we can then take it and attempt to give that customer a refund. We can bring together that capability that's probabilistic and deterministic with the agent and Claude here, and we can show that you can get that agent response right here. And our guardrails held. The customer's Wi-Fi was just out. Now that shows you how you can really bring together all of these capabilities, whether it be building agents, testing agents, and then distributing them across to other platforms right here inside of Salesforce and Claude. Now, thank you very much. Back to you, Christophe.

Awesome demo. A really, a really good example of software that can reason and act. Now, where users can experience that software is also changing. And to tell us how to build once and power every experience, Please welcome Principal Developer Advocate Alba Rivas.

Thanks, Christophe, and hi everyone. So Mohit and Stefan showed you what and how developers build in this new world, but where software shows up is also changing. You heard from Mark about AI Force, and what does AI-first mean for us developers? It means that now we can create custom experiences that show up in any interface, interfaces that didn't even exist until recently, like Claude, Slackbot, ChatGPT, Gemini. And we can do that in a way that's more open than ever. So in this chapter, I'm going to show you how to build those experiences with custom MCP servers, the headless experience layer, and multi-framework apps. And with those products, we are going to build a nice rollback firmware experience. So let's go to the demo, and please, a big round of applause for Rene and Sean, who are driving today. Thank you. So this is the firmware operations console that Mohit built. In production, it detected a faulty camera grid, and we need to roll it back. So let me step into the shoes of Leo, our operations manager. Like many people today, Leo works from an agent cloud, and he would like to roll back from there. But rolling back is Radiant's custom business logic. It's not something that you can do from the out-of-the-box Salesforce MCP servers. So that's a job for us developers. Radiant already had an Apex class. This Apex class handled firmware operations safely, and it enforces Radiant's custom rules, such as fixing safety-critical devices first. And you saw Stefan explain how to expose different tools through custom MCP servers. So that is exactly what we are going to do with this class. We are going to expose the Apex class as a tool in the server. This is a server that Radiant already had. There it is. And we're going to activate the server. So once that is done, we're going back to Cloud. And in Cloud, let's first make sure that the server is connected. And then let's ask it in plain language. Please roll back those cameras for me. So now under the hood, Claude is taking a look at the tools it has available. It's picking the right tool, which in this case is going to be the Apex class, and it's firing the rollback behind the scenes. So under the hood, It's safety, sorry, carefully fixing devices and picking safety-critical cameras first, which was the custom rule that Radiant already had. So this is how you expose your custom data and logic to agents. But the response was really a wall of text. So what if we give the agent a real UI. And we can do that thanks to the new headless experience layer. And we're going to start here in the headless experience layer playground. This is a learning tool where you can see your prebuilt widgets, you can see components, tutorials, and more. Also, you can create custom widgets and preview them here. And for the sake of time, we already Designed ours. But I'm not super happy with the title. So Renee, why don't you go ahead and change the size a little bit? There it is. I like it much better. So with HXL, you write the widget once using this declarative metadata UI format, and then it translates it into the language of the native target. For instance, Block Kit for Slack or MCPUI for Claude. And here in the playground, you can also preview how the widget looks in different interfaces, and you see that the styles change slightly. This is because it uses semantic variants that are rendered on runtime in the native look of each interface. So now let's continue from VS Code. This is the main JSON file for the widget. It defines its layout, its sections, its buttons, et cetera. It's exactly the same file that we saw in the playground. And then we need to create 2 nested custom Lightning types that are going to tell Salesforce how to render the Apex payload response with the widget. So now we are going back to Claude and we are going to try this out. We're going to make sure that the right MCP server is selected and we are going to ask it for an update. So we had to use a different MCP server because you need to set the Lightning type as output for the Apex tool so it knows that it needs to render the response with a widget and not with plain fields. And there we have the response, a real nice, rich interactive component rendered right here in Claude. Much better, right? So now we're going to try it out in another interface. We're going to Slack and we're going to ask exactly the same question. Please give me an update about the rollback. And again, Slackbot is taking a look at the tools it has available. It has the MCV tool because we preconnected it. And in a few seconds, you will see the response. It takes some seconds because it is real. And there it is, the same nice interactive card, but rendered right here in Slackbot. And with the styles adapted. To the interface. So one more test. Now we're going to try Agentforce. So this is the agent that Stefan created, and we're going to ask it for another update on the rollback. The configuration in this case is slightly different because it doesn't go through MCP, but the result is exactly the same. Here we have the same nice interactive widget. But one moment, because I said interactive multiple times, but I didn't actually show you, right? So let's try this out. Renee, why don't you go ahead and pause the rollback for me? Thank you. And there it is. So it's much better. Look how cool it is to be able to interact with the agent through clicks and not only through typing, typing, and typing. So this is HXL. You write the widget once and it renders in any interface and with the styling adapted. HXL is beta today, so you can try it right away after the keynote. And these are MCP apps. MCP apps are quickly becoming a standard on the web. And you can use them in Salesforce today. Great. So everything I showed you so far was surfaced through agents, but many people still work through the standard Salesforce UIs. And with Salesforce Multi- Framework, you can build experiences with the frameworks you love. So do you want to see an example of one?

Yes.

So let's go to the org and take a look at it. So Priya Nadar is one of Radiant's customers. She heard about the rollback, and she logged into Experience Cloud to know more. And here in Experience Cloud, she can see that one of her security cameras was flagged. The camera was off for some time and she can request a compensation. So she clicks on the button to get credit back. She's taken to this nice form where she can complete the information and request a refund. And then Stefan's refund agent will take it from there. So let's go to VS Code and take a look at how this site was written. Anyone here can tell me what this code is?

Angular.

Okay, it was not super challenging. It's written on the screen, okay? But it is super exciting because we started with React and now we have this full site written in Angular running natively on the multi-framework runtime and surfaced through Experience Cloud. So Radiant already has this side written, and they brought it into the platform as it was. And now it's running natively with authentication, governance, and security built in. And do you remember that Priya created a case? Okay, do you want to see the code for that? So let's open that second file in here. And look, this is GraphQL because until now GraphQL was only available for reading, but now you can use it for writing too with mutations. Mutations are GA and you can use them in your multi-framework apps and also in your Lightning Web Components. So let's go to the org for a second. because I want to show you one more thing, okay? This is the standard contact record page, a Lightning page. And on the left-hand side, we have some Lightning web components. But look at that component that we have on the right. Does it sound familiar? It's the same component that we had in the Angular side. And why is that? It's because it's still Angular. It's an Angular micro frontend. It's like an Angular mini app that's running here natively in the contact record page. And look, it's able to interact natively with the Lightning web components in the page thanks to the platform SDK and the Lightning message service. So these are multi-framework apps. We released them last year, we announced them last year at Dreamforce with React, and we keep making them better. Now having Angular, micro-frontends, and a ton of more features available in the Winter '27 release. And with that, let me close the chapter. This is how Radiance developers create custom experiences that show up in any interface. And in a way that's more open than ever. Thank you, and back to you, Christophe.

All right, thank you so much, Alba. I think that really shows the opportunity in front of us, create these custom experiences that show up anywhere in cloud, in Slack, anywhere, and reaching more users. So we showed you how, what, and where, and I hope you are as inspired as I am. But there is one developer in this room who is beyond inspired because he already embraced a lot of this innovation. Please give a warm Dreamforce welcome to the incredible Chatura Ranasinghe.

Chatura.

Thank you very much, Gustav.

Chatura, welcome to the developer keynote. You have been in the ecosystem for over 10 years. You have seen all the demos, So can you tell us how your role is evolving?

Yeah, actually, I have seen an incredible transformation of this amazing ecosystem and the community, especially over the past 10 years. And I started my career in 2016 while I was working at Burger King as a sandwich artist and also looking for different opportunities to build up my career on top of technology. So fortunately, one day I discovered Salesforce through a newspaper article. And that was one of my— actually, that was my main turning point in my life, the entire life. So after that, I decided to join the Trailblazer Community because I knew that's where the magic happens. And it did. It surely did. So at my very first meetup, I met my wife. And that was a main turning point as well.

She's here, right?

Yeah, she's also here. And after that, we led community groups in Sri Lanka. Now in Orlando, we both are running groups to help people and professional students to build up their career on top of this ecosystem and AI space. So as you said earlier, Christophe, yes, our role is expanding with AI in a meaningful way. So now it's time for us to build amazing things while focusing on our future developer skill sets in a proper way.

Awesome. Now, there are a lot of developers here who want to get started too. And they want concrete examples. So can you tell us a little bit about how you put AI to work on Salesforce in your company?

Actually, as an AI and computational neuroscience research scientist, I always wanted to bring my own AI models into Salesforce. I think I had a conversation with Mo as well a few years ago. And actually, the main reason I wanted to do that, I want to explore different areas in Salesforce. And I'm excited now because it's totally possible. And as a Salesforce developer, I always use coding agents to make my life easier. It's not only for coding. but also for turning my ideas into meaningful prototypes. And now with AI for and Dell 360 capabilities, I'm excited to explore how I can bring my data into Salesforce, especially into healthcare agentic workflows to streamline as well as to customize and give a personalized experience for the entire healthcare industry. And for me, it's time for us to get our hands dirty on what's possible and to see how it goes.

Awesome. Well, we can really see the impact of what you build, AI in healthcare. So thank you so much for everything you build, for the community as well. And before we let you go, we have a little token of appreciation for you. Please, everyone, join me in congratulating Satura on the Golden Hoodie.

Thank you very much.

Congratulations, Satura. It's well, well, well deserved. Thank you so much and congratulations.

Thank you very much.

All right.

All right. Congratulations, Chaturra. And now we can all be like Chaturra and embrace these new technologies and the new innovation. And to get you started, we are trailblazers. That's what we do. We put together an incredible series of virtual events and workshops. You can scan the QR code there or go directly to developer.salesforce.com. Thank you, Satish. We hope to see you there, but in the meantime, thank you so much for joining us today and enjoy the rest of Dreamforce. Thank you.

Thank you.