Architecture for the Agentic Future | Architect Keynote, Dreamforce 2026

Please welcome Senior Director, Admin and Architect Evangelism, Salesforce, Miriam McCabe.

How are we all feeling? Awesome. Welcome to the Architect Keynote. We're going to share our vision for architecture for the— for the agentic future, and that is something special. And can we just take a minute here and appreciate how big of a moment this is for all of us? Last Dreamforce, we relaunched a program. Now, this Dreamforce, we are here. Look at us. We're having an architect keynote, and we're having all of us here in the room to be a part of it. Yes, that is something. All right. Now we're going to talk about our vision for architecture for the agentic future. And before we get into the vision, I want to say a big thank you for all of you in the room for showing up and for everyone who embraced the program as we relaunched it. Your vision, your kind participation, your challenges, making sure that we were addressing them— that is what made this program really something we created together with the community. And that feels good. I am so thankful that you were there to do it with us. Now, this architect community is really something special. It amazes me every day. There is 63 architect community groups, 22 countries, All of that is built by you sharing your knowledge with one another. And all of it is part of the amazing Trailblazer Community that is celebrating its 20th anniversary this Dreamforce. That's right. Let's clap for that. Now, there are 51 of you in our very first cohort of Architect Ambassadors under the new program. And what you do is you amplify the reach of the program, making sure that our activities and resources are available to so many in the architect community. And you know what? You don't need to have architect in your job title to be a part of this community. If you do design work before you build, that's architecture, and you belong in this community too, and these resources—

Thank you.

are made for you as well. And let's have a look at this community. I've already seen some people sporting the Architect Ambassador t-shirt. Will you please stand up if you are an Architect Ambassador so we can see you? Amazing. Thank you so much. And can you please stay standing? No, let's, let's make this awkward. Please stay standing. And join them if you have ever been to an architect community group or have learned something from another architect. Can you please stand up if that's you? Oh, we need to do more knowledge sharing. All right, so now will you please stand up if you are here to share, grow, or learn? All right, that should get us going. All right. Now look around, look around. This is your community. Yeah, yeah, yeah, you can sit down. down. I'm not going to keep you standing the entire keynote. This is your community. This is the architect community, and it's never mattered more than it does today. And that is because architecture matters more than ever before. The enterprise landscape is expanding. We are bringing in agents as probabilistic actors. We are making all these new connectivities Through MCP and A2A, and we're bringing in large language models. It's a lot of moving pieces. And who is looking at the whole thing? Well, it's you. It's architects. This is what we do. This is what architects are made for. We know how to put all the pieces together. And we know that it takes a little bit of common sense to go along with the excitement. To make sure that our environments keep running, that our data stays secure, and that we also embrace the right pieces of innovation into our organization so that they can benefit from it. And that way of thinking, that is the value that we bring as architects. And we're doing it with the tools of the Salesforce platform. And you've probably seen this slide a few times already, this Dreamforce. If you look at all these layers, these products, these capabilities, all of them are the building blocks for the agentic enterprise. And nobody knows them better than you do. You take the decisions that allow organizations to bring these building blocks into their specific enterprise context. And that's not easy. In fact, that is getting harder and harder as the technology changes faster and faster, and you're expected to take those decisions faster and faster. Yes, it's a lot. But that is why it's so incredibly important to be able to rely on the trusted Well-Architected Framework. And this Dreamforce, we are announcing the biggest evolution of the framework to date, Introducing the new Well-Architected Framework. Yes, it's really something. It's 5 pillars designed for the agentic enterprise, helping architects navigate trade-offs and apply an agentic enterprise lens to the architecture designs. But there is more. It's built on the foundations of the Salesforce platform as well. And we know that resources, reading documents end-to-end, is not something that most people enjoy to do. And that's why we're making it easier to access all of that guidance. And that brings me to my second announcement today. We're also introducing the Architect Agent. It's—

Yes.

This is your AI-powered assistant. It lives on the Architecture Center, and it will give you answers to all of your architectural questions in an architecturally sound way. And it's grounded in all of the content, including the Well-Architected Framework, on the Architecture Center. Now, this type of evolution doesn't just happen overnight. We did this together with the community, and we're going to keep on evolving it as the community needs evolve and also as technology keeps on evolving. And so many of you helped us with your input and your feedback, and there is one person in this room that really went above and beyond. He's a technical architect. He is a community organizer. Please join me in welcoming to the stage Daniel Stange! Oh my God.

So good to be here.

We're matching. We have matching gold. I love the shoes. Can we have a round of applause for Daniel's shoes if you can see them? All right. Awesome, Daniel. So it's so good to have you here on the stage with us. And I want to ask you a few questions on your work on the Well-Architected Framework. Can you share what you think are the main challenges for architects and what that means for the framework?

Yeah. The first thing that is a big challenge for us as architects in this agentic era is that there's an abundance of things that we need to know about its products, its pricing models, you call it. Then we are asked for answers in and have to have answers in even shorter amounts of time. So we need to be really fast. And after all, The third big thing is that agentic actors, agentic stakeholders come into our workflows. So where there once was a lack of documentation of requirements, everything written, there's now an abundance of written stuff. And it is by agentic authors, by human authors, for agentic reviewers, for human reviewers, and about agentic systems. And I feel the Well-Architected Framework as we had it was relatively blank in that dimension, so we had to do something, I think.

Yes, and so I think a lot of these challenges will sound familiar to the folks here. How did it impact the expectations that you had for the framework? Can you say more about that?

Yeah, the framework as we had it was absolutely neatly designed. It had this impetus of asking ourselves, is it well architected, and then going into 3 dimensions and digging into 3 dimensions. And I was really keen to see this represented in a new framework in a way as well. And at the same time, the framework as it has evolved now has this kind of parallel structures of the agentic dimension and the human dimension, which is mirroring the— we still have the pillars augmented by the ones that are new for this specific time. So I think the outcome is really excellent and nice. And it helps us as humans and as architects to develop our vision, what is good. This is the thing that is guiding it. That's the kind of guidance that we're asked for all the time.

Yeah, and so I'm hearing you say that foundation that I was referring to earlier as well, right? That foundation of Salesforce best practice and then the agentic layer on top of that. I hear you. So do you have any advice for the folks here in the room that would like to use the Well-Architected Framework in their day-to-day, how to go about that?

Well, the first thing that I'd like to pick up always is a thought that my MVP colleague Shivanath has brought up. It is to develop your vision of what is good. And at times, good can also mean just good enough. So when we're working out and trying to solve a problem, try to comment on a problem and a requirement, we could start out and use this specific example to dig in.

Yeah.

You cannot really put all your org to a checklist and say, we check all the Well-Architected boxes one at a time. Well, we do have to check them one at a time whenever we pick up something. We can't do that for a whole org. But doing the exercise, asking yourself, is it well architected? And then applying the framework to the problem. This is, I'm really certain about it, giving better solutions. And this is the kind of oversight that we need. Thank you.

Absolutely. So I'm hearing you say don't boil the ocean, you know, take it one step at a time. And just for everybody here who's like, hmm, there is no checklist in the framework, there are checklists in the framework. So just to reassure those people. One last question for you. What is your favorite resource on the framework?

Okay. When I started out volunteering to do the review, The thing that I picked was my favorite pillar, which is the trust pillar. And I was really eager to see how the guidance is for specifically governance for agentic systems, because that is where nobody had a clear idea what exactly is coming at us. And at the same time, we are accountable for these kind of systems. And I feel governance is extremely important for that kind of thing. And I'm happy with the Dreamforce announcement in that area.

Absolutely. So trust is not just our number one value, it's also the foundation for good architecture, isn't it? Well, thank you so much for sharing your perspective with us here on the stage and also as part of the framework. There's more. I also want to thank you for everything that you do for the community as an Architect Ambassador. I'm going to save the list. You're amazing. You're such a role model. I am honored to present you With the golden hoodie, Daniel.

Woo!

I didn't see that coming.

You didn't see me coming?

You look amazing.

All right. Enjoy this moment, Daniel. Thank you so much. Let's have one more round of applause while Daniel gets back to his seat. Thank you, Daniel. Now let's continue. With our vision of architecture for the agentic future. We have 3 chapters as part of this architect keynote. We're going to start with Unlock Enterprise Innovation and see how organizations can bring AI interfaces into their organization in a responsible way. Then we will Scale Enterprise Impact and see how architects can take existing infrastructure and systems and get them ready for the agentic future. And last, we'll end on govern enterprise trust, and we'll see how the right governance processes will lead to the next best business opportunity. But let's start with unlock enterprise innovation. Please welcome on the stage distinguished architect, software engineering, Leo Tran.

All right, so we're seeing an undeniable shift in how users want to work. Users want to work with agents because agents help them get their work done faster. Now, the architectural challenge— responsible access. How do we enable users and agents to access just the data and actions they need without compromising enterprise security and accountability? And that's why AI Force Is the strategy for this moment. This is the strategy for every agent to be able to work with Salesforce on every interface, wherever agency users are. As Miriam said, architects are made for this moment. They're more important than ever because they're the ones that are going to translate this strategy into something that can be deployed to every enterprise. The architects are the ones that are going to have to figure out how to implement the key concepts of identity, connectivity, access controls. The Headless Toolkit is the accelerator for architects. It offers out-of-the-box capabilities for things like registering agentic identity, for connectivity through MCP, for additional access controls that downscope what an agent can do. The toolkit includes a headless experience layer so that agents can— and humans can interact more visually. It includes enhanced observability to differentiate when users and agents interact with Salesforce, and it includes prebuilt skills to accelerate how users can get quick time to value with agents. And with the recent partnership with Anthropic, it's easier than ever to connect Claude to Salesforce with the Salesforce and Claude plugin. And so to see how AI Force could be deployed to an enterprise, Let's bring up our demo company, Radiant Homes. Radiant Homes is a smart home services provider. They offer automation, security, wellness to residential and commercial properties. And it's renewal season. Residential plans are expiring, commercial contracts are up for negotiation, and their sales reps, they're racing to keep up, but that gap keeps widening. And so some of them have started turning to their own personal agent. They're not trying to go rogue, but they're sending a clear signal. They need the help of agents to be more productive. And so now it's up to the Radiant Homes architects to figure out how to implement, come up with an architecture for how to connect agents and their users with Salesforce. And to show you the challenge and the solution for how this could work, I'm going to step into the role of a Radiant Homes architect, along with 2 other fellow Radiant Homes architects, Lilith and Rajeev. Please give them a warm welcome. Lilith and Rajeev, we ready? You all ready? All right, let's do this. And so let's bring up onto the screen a personal agent that one of the Radiant Homes architects have looked to for help. And for those eagle-eyed watchers, you might notice, yep, it's the same agent they've been using to crack that world's best banana bread recipe The same agent they've been using to plan that 2-week vacation, and now it's going to be the agent to help them draft an email to a customer. And so they copied contact and contract information out of a report, gave it to the agent, asked it to draft an email, and it works, kind of. It's pretty generic. There's a lot more work to do, but we just witnessed 3 key enterprise risks. One, we just saw sensitive customer information now sitting in an ungoverned tool. 2, manual copy and pasting. We know it's fragile, brittle. It's a poor user experience. And 3, this agent only has a snippet of the data it needs. It's not going to be able to generate that personalized, high-value email that Radiant Home customers expect. And so this is the challenge. So how can Radiant Homes architects tackle this challenge? Well, first, I need to start coming up with a blueprint, an architecture. And so I can go to the brand new, because so much has changed, Well-Architected Framework at architect.salesforce.com. And trust is paramount. And so I could start at the trust pillar, and there's 27 minutes, to read. There's a lot of great content there. I'm an architect, I'm accountable, I will read this, but I'm ideating, and so I want to go faster. And now I have an agent. I can ask this agent for help. I can ask it, how can I connect users and agents to Salesforce securely? And it's processing through all of this new rich content, and it's giving me a blueprint. It's helping me think through access controls. the importance of zero data retention for compliance, how to think about human-in-the-loop oversight, privileged actions. This architect helps me go faster, or this agent helps me go faster. And so now I have the start of a blueprint. Let's get into the implementation details. And the first thing I'll start thinking about is identity. Radiant Homes has chosen to standardize on Claude. That's gonna be their agent for their sales reps, but how do we have Claude securely connect to Salesforce? Well, external client apps has been the way that architects and developers connect external systems to Salesforce. And Salesforce has done a lot of work to enhance this flow to enable registering agents. And so now, when Claude interacts with Salesforce, we know that it's an agent and not just another system. Next, connectivity. MCP is the de facto standard for how agents can successfully take action with systems. And Salesforce has enabled architects and developers to create custom MCP servers. Now, what I want to show you is something a little different. This is the Headless 360 MCP server. It is a single front door to the many capabilities of Salesforce. It simplifies connectivity, and it does so by offering 3 key tools. First, there's a discover tool. This helps agents go from intent to downselect to the relevant skills and actions it needs. Next, there's a describe tool to help an agent, given a skill, understand the request and response parameters. And then finally, a dispatch tool to help the agent actually take action. And this headless 360 MCP server, it's in open beta now. Go try it. Salesforce is going to continue to enhance this, And then GA'd in early November as the Salesforce MCP server. So, now we have connectivity. Let's talk about access controls. The typical pattern, when users and agents interact with Salesforce, the agent takes on all of the permissions and access that a user has. But that's not what users and architects really want. We want to restrict what an agent can do at first. That's where scoped access definitions come in. It enables a downscoping policy to restrict agents in addition to what users can do. And this works because remember how we registered Claude as a known agentic identity? Given that agentic identity, now we can associate a unique access control, scoped access definitions that persist across users. And so let's see what this, what this would look like for Radiant Homes. Here we have a scoped access definition that gives Claude access to all the contacts. 'Cause we want an agent to know and have context about the customers, but we don't want to give this agent access to create, edit, or delete to avoid unintentional changes to contacts. This is how you can implement progressive trust, how you can give an agent just the access it needs at the start, but have a paved path to enable an agent to do more over time. All right, so we have the foundation of identity, Connectivity, access controls. Let's now go back and see what the updated experience for the sales reps looks like. This is Claude. And now a sales rep has the Salesforce and Claude plugin ready for it. These are performance-tuned, optimized-for-outcome, prebuilt plugins available for the sales rep. And a sales rep can still interact with natural language with the agent, but these skills give a little more deterministic, quick way to achieve an outcome they want. And so we can try one of the skills, draft outreach. We can try and create an email to Casey Henderson at Meridian Property. And what Claude is now doing is it's using that connected MSP server, it's looking up contact information, it's looking up account history, it's now creating this personalized email. And what you're seeing here, this display widget, this is the headless experience layer, because we know blobs of text is hard for a human to quickly digest. What you're seeing here is a widget, the headless experience layer returning back a visually rich response to help humans understand what just happened. And this is a personalized email. If we zoom into it, we can now see Claude has figured out that Meridian Properties actually went from 36 to 42 devices in the past year. But with those 6 new devices, they missed adding monitoring. And so when we send this email to Casey, this is a great opportunity to remind Casey that he should be adding monitoring to his new devices. This is the help that sales reps want. And so let's recap. What did we just show? We showed how architects can turn eager but unvetted AI tools into trusted enterprise partners. They did this by building the foundation using the headless toolkit to register agents, have secure connectivity with MCP, and adequate access controls with scoped access definitions. And we showed this foundation with Clock, but this works with every agent, including the Salesforce-native ones in the Customer 360, Salesforce Coworker, Slackbot, as well as the growing ecosystem of third-party agents. And this foundation gives users the freedom to experiment and innovate and see how they can work better and faster with agents. That's why it's the start of enterprise innovation. So what's next? Now, Radiant Homes architects are ready to scale enterprise impact. And to help us learn more about how this thing happened, please give a warm welcome to lead technical architect, Shelley Souza.

I am so proud to be a member of this incredible architect community. I've never been more energized to talk about how architects bring value at scale. Let's talk about the toolkit we're bringing today to bring our next idea to life. So first, we'll talk about Data 360. That will help us read live data. Next, we're going to talk about Agent Fabric's MCP server. With that MCP server, we can unlock our APIs. Finally, we'll talk about AgentScript. AgentScript will help us apply logic for our next idea. So, what is our next idea? Well, let's go back to Radiant Homes. So, Radiant Homes is in the middle of renewal season, and it's busy right now, but they have a fast, sanctioned way to get grounded answers on their individual accounts. So, we want to think through what's the next step to take our idea to the next level. Well, what if the agent could create a renewal proposal automatically. Well, we don't have to start from scratch, and we shouldn't. We already have data in IronCloud— that's our contract management system. We also have smart device data in Snowflake. So we want to pull all of that together, but how? Let's hop over to our Architect agent. So I have an existing agent, And I want to bring it into an environment with an existing Snowflake instance and an existing integration layer. What should I consider? The agent pulls together all of the key decisions that we need to think through. The very first one, Data360 zero-copy federation, because we want to be able to access data, live telemetry data. We want to do that quickly, but we don't want to duplicate it anywhere. The next, we need to think about how we're going to access these APIs. from our contract management system. We need a transactional, multi-step scenario in an API layer that's governed. We also want retries, and we want policy enforcement throughout. Next, it's really, really clear on governance, least privilege access, and full auditability of every agent action. Okay, so my wonderful demo partner Lilith and I, we're going to put on our architect We're going to pop them on. They're very, very agentic, very architecty. Yes. And we're going to take a page out of our Think Like an Architect playbook. Is anybody using the Think Like an Architect playbook, that series? I hope you all have seen it. It's fantastic. I see some hands. So, what we want to do is break this down into a few high-level requirements that will make it easier for us to walk through this design. So, the very first requirement is make telemetry data agent-ready. The second, bring in contract terms from Ironclad. And the third, consistent agent reasoning across every surface. Okay, first requirement, make telemetry data agent-ready. So Lilith, help us out here. What's the first assumption we're making? I think that assumption is our data is already in Snowflake, and that's true. So all of our smart device telemetry data exists in Snowflake today. And we have to ask ourselves a question because, well, architects ask ourselves questions and everybody else many questions as well, right? Should the data live in Snowflake? And the answer is yes. It's not a trick question because this is smart device data, so we want to be able to access it immediately, right away, as soon as it streams in. Well, that decision is easy. From there, we'll go to Data360 and we'll scope it with zero copy. So let's hop over to our demo environment, take a closer look. So you can see here we have a connection already created. It's Snowflake with zero copy. We don't have to copy any actual data. This isn't like ETL. We're accessing this data live. Let's hop over to our query editor. So when the query editor loads, what we'll do is we'll click to run a live query. We'll check the data, we'll go out to Snowflake, grab it immediately, and we didn't need a big migration project to to make this work. It's accessing it right now. Okay, next requirement. We need to bring in contract terms from Ironclad. So, another assumption we're making. We brought in all of our B2B accounts to Ironclad. We migrated those during an acquisition years ago. So, we need to decide how we can access them through the APIs. Of course, APIs exist already, right? Ironclad has them. We use them every day. But how do we make them accessible to an agent? So we need to figure out a way to have the agent understand what this API does. So we're going to expose this API to the agent as tools. Agent Fabric to the rescue with MCP Bridge. Next step, we'll wrap that API in a custom MCP server. We'll expose that to the agent as tools, and then it will be able to understand how to read that data. So Let's hop over to our MuleSoft environment and take a closer look. You can see here, now with our API wrapped in an MCP server, it has tools like get contracts and update contract. It'll be able to understand what to do with this data and create that renewal proposal behind the scenes. All right, last requirement. We need consistent agent reasoning across all surfaces. So, big question we have to ask ourselves here. Where will we build this agent? We know that our reps like to use different surfaces for engagement. They can use Claude, Slack, or Salesforce. Any of these work for them, and we want them to work where they work best. It's also really important that we insert some probabilistic and deterministic logic, because we want to make sure that the agent is able to be creative. We don't want boring emails, but we don't want it to kind of go off the rails and do whatever it wants. We want to give it some rules. So, with that in mind, we'll create an agent force agent behind the scenes, but we'll also use agent script, and that'll apply our logic. Okay, let's take a look at this in the actual environment where our reps are working today, and that's in Claude. Just like before with Leo, we will ask for a renewal proposal for Casey Henderson ahead of Meridian Towers renewal. And we're going to give this a moment to cook. And I say cook because unlike a cooking show where we have a pre-baked cake that's in an oven over here somewhere, we just pop it out. We're baking this thing live in front of all of you. So, it's going to take a minute to pull everything together. But it's working, and we're already getting information we didn't have before. We didn't have upsell opportunities. We didn't know that we needed to replace hardware, but it's telling us this now. Next step, we need to pull together some of the records behind the scenes in order to pull together this renewal proposal. So, we'll create some opportunities, line items, price books, We'll hook all of those together. And you can think, you know, a rep would've been sitting here maybe copying and pasting, pulling that data from different screens. They don't have to enter this now. It's being done for them. What would've taken them potentially hours now takes minutes. Okay, last step. We're going to draft the actual renewal proposal. We want to get this out to our reps and let them see it. So, the renewal proposal is being pulled together. And again, you know, you all are watching something live. You know, you know the difference between a live demo and a slide deck, right? This is happening for real. But think again. This happened right in front of you. It took seconds, and now we have a real proposal all grounded in our Snowflake data and our Ironclad data ready to go. So the point here is Radiant Homes did not have to throw out years of infrastructure to go agentic. They took what they already had and they made it agentic compatible. So Let's go back to this slide. Let's talk a little bit about these products again, because we took these products, right? We took all of these items, and we pulled them together into a new vision where we have a new agent that's live in production. Our reps are using it. They're moving a lot faster today, and they're giving better quality to their customers. And that's how you scale enterprise impact as an architect.

Woo!

And now, to teach us how to govern enterprise trust, one of the amazing architects behind our new Well-Architected Framework, the wonderful software engineering architect, Tim Breeden.

Hello, everybody.

Wow.

Take all you in for a second. I promised myself I would do that. You're all out in the dark and there's bright lights. And, um, the reason I promised myself I would do that is there are a lot of people sitting out, uh, in the audience that I'll get to meet in person for the first time at Dreamforce. And that's really exciting for me. Um, this has been over the past 6 months being so focused on the new framework. A lot of people in here have contributed to that. It has been a community effort, so I have to give a sincere thank you, and you know who you are, to the people who provided what I would call unfiltered, unyielding feedback. And even when it went to Miriam, it went straight to me like a vacuum tube. So Thank you. That, to me, the tough love especially, that's community. And this is a framework created by the community, not just for the community. Thank you. All right, but this is just one big step. This is an evolution of the framework that we're very excited about, but this collaboration has to keep going. We have to keep it up, do this together, keep evolving it. and keep it working. So speaking of keeping it working, that is exactly what I'm here to talk about. So getting an agent up and running, that's hardly the finish line, right? You have to stay close to it. You have to keep it trusted. You have to listen to what it's telling you. And that's governance. And you do governance right, it doesn't just keep your agent healthy. It's how you can grow. It's how you can discover The next opportunity. So how do you go about doing that? Well, let's go through things at a high level real quick before we go back and we dig into what's going on at Radiant Homes. So what we'll do is we'll start at the architecture center, and you'll get to see that agent again. Then we'll look at agent force observability and see how this agent is behaving in the wild. And for consumption and usage information, we'll We'll check out Digital Wallet, and then we'll take signals from all of this and feed it into Slack with Tableau Next, so that the entire team, the center of excellence we're gonna talk about, can work on things, and communicate on things, and stay in sync on things in real time. And here is the thread I want you to keep in your mind as we walk through this. All of this is about governing what's already up and running. And all of it's about discovering the next opportunity. So if there's, you know, a phrase in mind, you govern what's running and you discover what's next. So back to Radiant Homes. We've got our agent, it's up and running. Wonderful. Our first agent force agent. Okay, there's no argument we need to govern this thing. We get it. This is serious stuff. But are we ready to? We don't know. So how do we find out? Well, let's not take a guess. Let's go to where the guidance is. Lilith, I think you know where that is. Let's go to the Architecture Center and ask the architecture agent exactly what's on our mind. How do we know if our center of excellence is ready to govern agentic workloads? And I love it that it's grounded. How many times have you heard grounded so far?

3.

in this presentation. It is grounded in the framework. Right out of the gate, it points us at the right tools for the job, the ones I just mentioned: Agent Force Observability and Digital Wallet. Scroll down a little bit more. Your COE should enforce least privilege access. I think I see that phrase more than anything else— least privilege access. So there's that trust pillar. Daniel, thank you. It recommends an accelerated communication and sync cadence for governance, which we'll see in our center of excellence. So you gotta move fast 'cause the agentic world is moving fast and your agent can do a whole lot of things real fast that you need to govern as well. And finally, a cool little checklist. Data quality baseline, tech debt register, sign off on schema changes, And a documented AI governance policy. And that one we don't have just yet at Radiant Homes. But we've got an answer, and we got our first gap to fill. So how do we go about filling that gap? Well, what we do is we start putting into practice the recommendations that the agent gave us, and that starts with how the Center of Excellence is going to work. So easy enough, we set up a, you know, a weekly standup, a sync Where our developers and our admins and our architects, especially the architects, get together, stay on the same page, make sure that the way we're governing things and the changes we're making aren't conflicting with agents that are already up and in the wild. But that accelerated sync that got recommended to us, what we have is just the tool for that, don't we, Lilith? A Slack channel, always on, where the team can stay in sync real time, all the time. Something comes up, send a note to the channel, taking the signals from those other tools and working on it together right in the channel, taking action. And even as I speak at the Architect Keynote, the team is in the channel working through changes.

Good stuff.

All right, so we've got our process updated. It's in place. Now, I'm an architect, our first Agentforce agent ever. I really wanna see how this thing is behaving in the wild. So Lilith hops in through Agentforce Studio down to that sessions and intents tab, and we look at Agentforce observability. Now, I know sometimes, Lilith, you like to get the right fields in. the wrong fields out so that you see the things you want to see and that, you know, we both need to see as architects how this agent's doing. Beautiful custom view set up. So we start looking at it. All right, first agent, first agent. Start scrolling through and things look pretty healthy at first, but then there's an error. And it's an error, it's not a heads up. Okay? That's why it is Red. So we need to dig into it, and we go in and we go all the way to the trace and see what we discover. And it's an interesting thing that's happened here. The rep did ask for a proposal, and the rep did get the renewal proposal, but it was very generic, very one-size-fits-all. It didn't really say anything about this particular customer. So that's a partially grounded suggestion. And the cool thing is that the agent really flagged its own work. So there was data already in Radiant Homes' own records mentioned earlier by Shelley. This is device usage consumption information that the agent was not yet grounded in. So what happens is you get a proposal any customer could have gotten, and that's not our best work. And that's what makes it an error, not a heads up, because there's a standard, Radiant Homes. wants to live up to. So now, cool, we found a weak spot and the agent pointed us toward the fix, and we're going to remedy that and get that agent grounded in the right data. But I'm also interested in cost and consumption, and in particular for this being our first agent force agent, because it's going to form sort of a de facto baseline. So Will you pull up Digital Wallet? Thank you. It will establish a de facto baseline that we're going to compare future agents against. This is our first time through it. So we go in and we look at consumption by agent, our one agent. And as an architect new to deploying these things in the wild, it looks good. I mean, it looks busy. The resource and cost optimization pillar warns us about runaway busy. that just eats up credits faster than any human could. But this looks healthy busy to me. So I'm feeling pretty good. Take stock of everything. Quality, we found the weak spot. We fixed it. Consumption, yeah, I think so. I mean, I think it looks good, and our reps are really using this thing. But I'm trying to figure out what to make of all of this because I'm not just entirely sure. I don't have the experience yet. So I think to myself, Who is the one person that I know, that we all know, who likes to just tinker on problems like this, figure out the direction something's going? Lilith van Beeson.

Yes.

Lilith, what do you make of all this?

Yes, Tim, thank you. This is really, really interesting. I'm very happy that you shared this because now we have a better view on how our agent is how it is consuming, but what we don't have a view on is the outcome. Is it really bringing that result that we were hoping for, that increased renewal rate? So where else do I go than to my best friend, my big buddy, Slackbot? And I'm gonna ask it to just highlight this for me. Tell me exactly how these insights work together so that I can see if we are getting that increased renewal rate. And I'm going to put Slackbot to work, but I'm not going to make you wait because this is— I am the kind of cooking show that is already baking. So I did my baking this morning. And here we have the beautiful visualizations that we get because Slackbot is connected through the trusted semantics of Tableau Next MCP. And this means that I get an immediate answer to my question. Is there an increased renewal rate for the agent-guided versus no agent involvement? Well, clearly, yes, we see that our agent is actually doing what it's supposed to do. But it gets even more interesting because we see that for those contracts that have been renewed and where we had an agent guiding it, we have more products being added. So our agent is not helping only with the renewal rate, it's also helping to upsell and increase the usage of these products for our customers, which is very interesting. So I don't want to sit on this information alone. I want to share this with my team. So let me bring this to my COE channel and tell them to take a look. And what is really, really cool is because I'm bringing this there, they can continue this conversation. They can dive in deeper with Slack routing channels. This means that everyone can continue and go deeper and yeah, get the right insights right at their fingertips. Isn't that interesting, Tim?

Very. Um, and thank you so much for getting the note sent off to the team. I appreciate that. But I want to make sure that I got you clear before we get back to all these nice people. So the accounts that this agent managed, we saw a 90% renewal rate. Is that right?

Yes.

And compared to the accounts that the agent does not manage, we saw about a 4x addition in products and services. from the year before, right? Wow. So you think about what governance just did here. It's the same guidance driving both loyalty and growth. So kind of maybe the obvious next thought is, why are we even waiting on renewal season at all to take action? You know, Leo had his reps running around like that because it was renewal season. Well, that doesn't have to be the case, and they can be even more productive. If we have an agent that can watch all of this data all the time. It could flag us in real time, you know, if an account is ready for a tier upgrade, if an account is quietly falling off, and we can do something about it before it turns into a red account. So this to me is amazing. Governance done right is not just about protection, it's how you grow. So govern what's running, discover what's next, and that's how you govern enterprise trust. Miriam, will you come close it out?

And can we just have another round of applause for these amazing architects, the presenters, and the demo drivers? We just watched these architects share their vision for architecture for the agentic future. We saw Leo unlock enterprise innovation. We saw Shelly scale enterprise impact. And we saw Tim and Lilith govern enterprise trust. All of this was made possible by architects. Architects taking decisions, architects thinking about trade-offs, and architects applying and interpreting the knowledge and the answers the Well-Architected Framework provided. Architects unlock, scale, and govern the agentic enterprise and the agentic future, and that future is now. And here's what's next. Please help us shape the future of the framework by providing your feedback. Scan the QR code. Use the framework in your day-to-day so that we can use this to make it better in the next iteration. And with that, we have come to the end of this Architect Keynote at Dreamforce. We really hope you enjoyed it. Thank you for your time, for sticking around. Enjoy the rest of Dreamforce. Thank you.