Become an Agentic Enterprise | Agentforce Keynote, Dreamforce 2025

Please welcome Senior Director, Product Marketing, Salesforce, Ryan Shellack.

All right, welcome, welcome.

Day two of Dreamforce, we're doing this. We got Dreamfest tonight. Metallica and or Benson Boone. You can probably do both. You can do both. And hey, we want to thank you for joining us for the AgentForce keynote. And this is gonna be a great party. We have a ton of good content, products you've never seen before. But we're gonna get it started, right? Cause we're gonna start it off with all of you and our wonderful community. Our Agent Blazer community is 90,000 strong on Slack, and we really appreciate what you all are doing. We're building this product with you and for you. You're using this community to share what you're learning. You're finding use cases with each other and you're supporting each other in your journey to be Agent Force experts. I want to take a moment over here to talk to somebody who really exemplifies what this community is all about. So I'm going to invite up Joey Monroe from Blue Gator. Let's all welcome Joey to the stage. Joey is AgentForce OG I know the product's only been around for 11 months, but. Joey, good to see you, brother. How are you doing?

Doing well, thank you very much, Joey.

I mean, you started last year. You didn't know anything about AI agents. You built your expertise. I want to ask you, what has this community and trailhead meant to you and as you've become an Agent Force expert at your company? Yeah. So the Agent Agent Blazer Trails have been so instrumental in being able to kind of logically move through, learning everything about Agent Force and really building on that fundamental AI structure. And the Agent Blazer community has really meant so much to me because it was a way to connect with others that were learning the same things that I was doing, running into the same hurdles that I was doing and just bouncing ideas off of each other. And that that community has grown so large and so, so great that it's just been awesome to be a part of. Yeah, you've been a major contributor publicly in that community, but also, how have you been internally supporting your colleagues, your buddies, as they've also become Agent Forest experts within bluegator. Right. So within our own organization, I kind of became the subject matter expert because I was the first one to take the leap and to be able to do that and then start to learn the lessons. I was able to then create workshops and create study groups to get the rest of our consultants certified. Every last one of them is now AgentForce certified and we've had such a great adoption. Everyone's super excited about the product. Now I'm in a position to where it feels like I can give back and I know a little bit more than I did before and it's just confirming that. That's awesome. And hey, we're going to let you go in a second over here. But before we do that, we want to honor you for all of your contributions to this community by awarding you the first ever golden hoodie at an Agent Force keynote. Come on up here. Congratulations, Joey. Thank you so much. I'll take that for you.

Woo.

Yeah.

Let's get a picture over here. This is looking.

All right.

Now. You ready to get this keynote started? We got a huge party for you. Let's start off with a little hype reel of everything we've been doing building agents together for a year. Let's roll that build.

Agents aren't just code.

It controls the chaos that emerges from.

You think building agents is easy.

It's layers of complexity.

Agent Force is fundamentally changing the nature of the digital world. This is the Internet again. This is electricity.

It's that level of change.

I was able to build an agent so easy and fast. And more importantly, it actually worked.

It doesn't require a lot of coding.

I can just simply ask it a.

Question and it gives me the response.

AgentForce can reason and can make real time proactive recommendations.

It really is like we all have a bunch of assistants working on our behalf.

AgentForce voice allows us to create efficiency in the overall service experience so that.

Our service representatives can focus on the.

Things that need human attention. Our Salesforce agents can work with agents from other platforms. It's going to make it so that we can drive innovation in a much more impactful way. Magentix solutions allow us to scale some of the things that we want to do at an exponential rate. Within the first few days, we saw.

A 60% improvement in deflection rate.

We saw 185% increase in number of appointments. We have seen a massive efficiency where we have implemented AgentForce in the range.

Of 25 to 30%.

In reality, it amplifies the human.

That's the power of a human plus a digital worker.

Partnering with Salesforce with agentforce at the core is how we make the future work for everyone.

Please welcome executive vice president and general manager Salesforce AI platform, Adam Evans.

All right, all right.

Welcome to day two of Dreamforce and year two of Agent Force. It is amazing that a year ago, right here we launched Agentforce and we have been busy. There is a ton of product innovation that we're going to go through and we're going to dig deep. We're going to do it through the lens of some of our amazing customers here today. And we're going to tell you about all the lessons that we've learned to build great agents that work. And we're going to do that in the next 45 minutes. But first I want to say thank you, thank you to the 12,000 AgentForce customers making AgentForce the fastest growing product in Salesforce's history. Companies like Williams Sonoma, great retailers in Latin America like Vallabella, serving their 35 million customers on WhatsApp. Customers like Indeed and Adecco, connecting employers and job seekers to create new jobs and opportunities with AgentForce. And fast growing upstarts like Engine that are serving up to a million travelers helping them find accommodations in Europe with Agentforce. It's incredible and you've probably heard the story of help.salesforce.com, our customer zero story, 1.8 million conversations, 77% resolution rate, tremendous operational impact for Salesforce. However, it didn't start that way. We're going to talk a little bit about the learnings of making great agents. In fact, when this first started, the resolution weight was about half of that. It was through continued investment, through management, observability, tuning the prompts and the agents and context and getting better data that we got this kind of performance. We've all heard the stat. 95% of pilots never make it to production. There's a lot there because prompts and a loop are not enough. You need a platform behind your agent. You need to be able to connect that agent to all the channels where your employees live like Slack or your customers live on your website or your phone. Now your contact center with agent Force Voice. You need to have the control with a reasoning engine for your most complex business process that creates the most value. You need to be able to have the tools to develop agents as the new apps with building, testing, deploying, version control and iteration to get the results you want. And of course trust the guardrails that work, the fine tuned models, ultra low latency that keep your agents performing with the policies that they should be working with and context. You need data, whether that's through zero copy for massive amounts of data that you're bringing in or in an open ecosystem with MCP and a to a. You want to connect it all. And we have been busy creating the latest version of Agent Force with the most tools to unlock all of this. And I'm very excited to announce Agent Force 360. AgentForce 360. Yeah, we can clap Agent Force 360. We're introducing more control than ever. We're opening up the reasoning engine for you with agentscript. For some jobs when you want control and other times you want creativity, you can have it all. We're making it more accessible than ever with a brand new agent builder built from the ground up that's clean, beautiful, and it brings your whole team along to build agents together. And of course, unstructured data continues to get more and more complex and you need to have new tools with intelligent context processing to let your agents understand that data and bring it in at the right moment and create that accuracy. And all of these things are also now available, not just in chat, but your agents can talk with AgentForce Voice. We couldn't be more excited. So this isn't just about all the features and the new innovations that we've had. This is also about all the learnings that we've had working with you about how to create great agents that really work. And it boils down to four major components. One is being able to create agents that perform with the control and consistency that you want for the job at hand. Two is about being able to pull in the right data, having the right context and bringing it in to create the personal and great experience that you want. Three is being able to take that data all the way to the user, whether it's in chat with rich carousels and photos and images, or connect to the customer where they are in your contact center with voice. And it doesn't stop with just building agents with data and launching them with great experiences. That's the beginning. You need tools to manage the observability, the tuning, the refinement. Just like help.salesforce.com you need all of this to create great agents. And we're going to break it down for you. So let's jump in. I'm going to hand it over to Nancy Hsu to get started. Nancy.

Thanks, Adam. We all know AI is moving fast. But you know what's even more exciting? Who is building AI is rapidly changing. It's no longer just your admins or your developers, it's your line of business leaders, your IT teams, your service teams, your sales teams. Every role and every function needs to become an agent builder. Now here's the thing. You don't just need agents that are adaptive and dynamic. We also know now that you need Agents that are predictable and controllable. Because when your agent goes off key during your VIP conversations, unpredictability isn't charming, it's costly. The reality though, is that most agent platforms today don't give you both of these things. They're either too rigid. So when you have a retirement agent, it can't answer basic questions about your retirement accounts because it doesn't have that logic hard coded into your agents, or they're too open ended. Meaning your agents, yes, they will do that retirement transfer for you, but they're going to do it in a way that doesn't follow your business processes. You need your agent to be able to do both of these things for you consistently, predictably, every single time. This is why we are so excited by the brand new AgentForce Builder. The brand new AgentForce Builder is making it possible for anybody to build an agent. To do that, we are now giving you three new easy ways to build. You can author in natural language, you can get this vibe, code your agents conversationally, or if you're technical like me, we are going to make it possible for you to go straight to agentscript, our brand new scripting language that is making it possible for anybody to build an agent that is, yes, powerful but also predictable. I like Formula One and the analogy that I like to give for this is we're making your agents into Formula One drivers. We're going to get you to your destination, we're going to adapt to the road conditions, but most importantly, we are going to make sure your agents stay on track for you every single time. Today we have thousands of our customers, all of you in this audience today, and I'm really excited that we have Ishwar Vazori and the CTO of Equinox. Ishwar, come on stage. Ishwar, it's so great to see you.

Good to see you.

You're here at Dreamforce. How do you feel?

Very good. Excited and even more exciting to see all the new things that you're launching.

We're so excited to have you here, Ishwar. And one of the really exciting things about what you're doing at Equinox is not only are you building an agent with us at AgentForce, your agent is now live on your website. So I thought today why don't we go ahead and just show the audience here what you built with agentforce and let's talk through some of your learnings.

Definitely, let's do that.

All right.

This is a live demo and I think let me first make sure that we Have a good use case that we can walk you through. So are you a member of Equinox right now?

I'm not a member though, Ishwar. I just moved to San Francisco and I'm looking for a great club.

Okay. So that's a perfect use case where anyone coming to our website, we call them prospects, they want to know which clubs are near me. Right. So let's try that out.

Let's do it.

So here I'm asking which clubs are near where we are. And it basically provides a list of clubs along with the basic information regarding the amenities, the address, and including the images of the club that were rendered kind of from our website. One of the things that is important here to note is this is not coming from a knowledge base, but it's connected to our API so that the business logic that you were talking about just before is incorporated into the chat and the answers that it's providing. So now when you want to basically take a class or something, what time would you be interested in doing that?

Yeah, this is really powerful, Yishua. I know. I walked by a couple of these gyms before. Before I like to work out, let's say 5pm in the afternoon before dinner.

All right.

So this is another usual set of questions that our prospects have, because the thing that they enjoy about Equinox are the classes, and people want to know all the offerings and whether that works out for their schedule and so on. So here we're able to show you that you have a Vinyasa yoga, you have an anthem ride, and these are the classes that are available so that you understand the offerings and make your decisions based on that.

That's awesome. I love this. Eswar, you've sold me. I'm going to become an Equinox customer thanks to you.

I'm going to count on that.

I'll see you at the club. Not only that, because your agent is live, I can do that entire process. We just went through live after this session to become a customer, so that's really exciting. Now, Ishwar, before I let you go here, we were just talking about some of the learnings we've had in AgentForce over the last year. What are some of the learnings you've had working with us at Equinox to enable your idea of the agentic enterprise?

I think we took some of the brand tone, how to get the brand tone right, how to get the responses right. When do you need to be terse? When do you need to provide the information? I think that's Just a process of training the agent to kind of depending on the situation, but also understanding the personality of the user. And perhaps saying that Nancy requires a longer response, whereas John might require a terser response. Right. And the second one is really around the feedback loop, making sure that we have a team who tells the agent when it did a good job or when it did not do a good job. Because if that feedback were not provided, then the learning that we are counting on does not happen at the pace that we would want.

The feedback is very important. What does the future look like for you as you think about our partnership together?

We're working on several fronts, but one of the main areas that we're really excited about is to be able to provide coaching information to our members. One of the things that has always been challenging is when you provide a recommendation. Let's imagine that you went for a long run and we gave you some feedback. You want to actually tweak it a little bit and that was not possible without having a conversational ability or a nutrition tip. And you say, no, I don't like protein, I like this. So the ability for an agent to take in your input, modify the recommendation, but also remember your preferences for the next time and provide that contextual, anticipatory information is something that we are working on and we are very excited about.

Thank you Ishwar and thank you for such a great partnership. We're so excited help continuing to enabling that vision. Ishwar, you have convinced us how important a health coach agent is. And in order to show you that, I would like to invite up Aaron Kahl, one of our incredible product leaders on the AgentForce team who has been leading this effort to build the brand new builder. Aaron, take it away.

Awesome. Thank you, Nancy. It has been incredible to see the success that customers like Equinox have had in the year since we launched Agent Force. But we've also heard you that you need more. You need tools that give you full control over your agent's behavior because you want agents to power your most mission critical use cases in production at scale. Well, we're about to show you those tools today. Let's dive right into the demo. You just heard how Equinox wants to build AI health coach agents that scale the expertise of their trainers to every member at every club worldwide. But here is the challenge. When you start to translate this use case to an agent, you end up with instructions that look like this. Before this, do that, but always do this. Unless this happens, then do that. And the more of these Instructions you write, the more they start to overlap. And the more they overlap, the less consistent your agent becomes. And you find yourself playing prompt, whack a mole, testing and tuning instructions, trying to persuade an LLM to do what you want it to do every time. Now, we have worked with thousands of customers over the last year and we know this challenge is real. It is frustrating. And if you deal with sensitive information like PII or financial data, it is a deal breaker to going live. And you need a better way. Let's meet the new AgentForce builder. The first thing you might notice is this looks. Like we have radically simplified the interface and given you a clean canvas where defining your agent's behavior is as smooth as writing a document. Just like a document, we can start by pasting in a few instructions. And even though we just started, that familiar challenge has already emerged. But watch this. Let's ask AgentForce to enhance that instruction so the agent can execute it more reliably. Look what happened here. We were able to break down that natural language into rules and variables, executable logic. Your agent can run the same way every single time. That is blended in line with your natural language instructions. That means that you have the best of both worlds. You have fluid agentic reasoning and you have rules based control in one instruction set. Now, the breakthrough that makes this possible is Agent Scripts. AgentScript is our new scripting language that we built to unlock the patterns you need for those mission critical use cases. Determinist chaining actions, conditional logic, collecting data. If you're an admin or a line of business builder, you can take advantage of all of that in this canvas. In a guided document like Experience, if you're a developer, you can switch to the script view where you can directly edit the script for even more precision and to build with more speed. And everything you do in either view is always in sync. Now, AgentScript isn't just sitting behind a single topic. Your entire agent is represented in AgentScript. What does that mean? It means it's copy pasteable. You can share it, it's version controlled. It is a modern way to build that is gonna help you build more powerful agents than ever before. And we just used it to give our health coach agent an upgrade. And we all know the next step is to test. To help us do that in our upgraded preview experience, I'd like to introduce my partner building this project, Angela Lee.

All right, Erin. Well, this new simulator experience is going.

To pump you up.

The simulator is where you're going to test and iterate your agent throughout that building process, let's take a look at how it works. So let's ask the health Coach agent to create a workout for me. And in the middle you see the steps that the Atlas Reasoning engine is taking to reason through that request. And we also get this really nice, easy to understand summary of what just happened. But if you're like me and you like to sweat the details, you can now dive really deep and see a full trace of everything that the agent did to get to this response. And why is this important? Because when AI reasoning is a black box, it makes it really difficult for our customers to be able to figure out what happened and then how to fix it. Okay, so this response looks okay, but it's actually treating all of our members the same way. If I'm a member who is on a trial period, I actually don't have access to in person classes. So this recommendation actually doesn't make a lot of sense for me. This is where a lot of our customers got stuck in the last year. They see a response, they know what needs to be changed, but they don't know what lever to pull in order to make sure that they're tuning the agent. Well, now we have AgentForce. Let's ask AgentForce to help us make sure that the Health Coach agent is always checking for membership type before providing a recommendation. As you can see, it recommended that we add some logic right at the start of the beginning of these instructions before the agent gets to work. Without this if then statement, we actually would be writing another long text instruction that would be sent to the LLM and then we really hope that it figures it out. But now we have these conditional controls and that context engineering that ensures that our Health Coach agent is always going to ask for that membership status before providing a recommendation every single time. All right, Erin, that was a little bit about the simulator. What other heavy lifting can this agent do?

Awesome. Thank you so much. Angela. You've seen how you can build in the canvas. You can edit the script directly, you can build conversationally, which is amazing. But there's still one more thing because if you look closely at this topic, it's not just helping plan workouts, it's also saying when to transition to the local club recommendations topic. You can now orchestrate across your topics as sub agents. You can build a network of agents to extend your agents like never before. And because AgentForce is open and interoperable, it is not just AgentForce agents. You can connect to third party agents with protocols like MCP and 8A and for Equinox, they've partnered with Escape to provide robot massages to their members to help them recover from a particularly intense workout. Awesome. Our agent is looking great. I think it's time to see it in action. So we'll activate it, then let's head over to the Equinox app and see it in action. We'll see. We can start a conversation with the agent, we'll ask for a workout plan and we'll see. The agent is able to provide a personalized workout plan in this rich, beautifully formatted response, and even knows I'm not.

A member quite yet.

So thanks to our instructions, it recommends a tour I can take at a local club. And this is just fantastic. And for all of you who have been building agents, it is what you have been waiting for. It is the magic of agentic reasoning with the certainty of rules based control in one modern interface. And we cannot wait to see what you build with it. Next up, Madhav.

Great job, Aaron. All right, so we talked a lot about control. We're going to spend a little time and talk about data. Now, if you listen to an AI podcast, some of you in the audience, you know who you are, some AI thought leadership texts, threads on X and LinkedIn, you're always going to talk about bad data equals bad AI, right? Garbage in, garbage out. Now, we've been very fortunate to work with thousands of you over the past year to really dial in what is the data problem. We're going to share some of that with you today. So the first problem that agents have is to try and understand vast amounts of different kinds of data. Now, some things are easy. I have a text file, I've got structured information. It's easy for an agent to understand and interpret that. But remember, your content is the accumulated wisdom of everything you have built in your company. And that content takes a lot of forms, rich formats, pictures, tables, workflows. Agents struggle to understand all this information without a little bit of help. So that's problem number one. Problem number two is a little more subtle, but quite profound. What the agent uses the data for really matters. Now imagine I've got a rich document that describes my product. That same document could be used by a seller to sell the value of the product. It could be used by someone in service for a completely different purpose, completely different context. So when you add those two things up, what we have to do is build tools that allow you to turn content into context, helping you ensure we can ingest, interpret and index all of this great information that you have spent years curating for agents to use them to drive that accurate response that drives the outcome that you care about. All right, so now let's see how this works. And to show us, I want to welcome Alison, everybody. Give Alison a big hand. So in the keynote yesterday, you saw the incredible experience from William Sonoma. It's rich. It's got recipes, it's got product information. All of this comes from the incredibly rich content that Williams Sonoma and all of their partners and companies and, and providers have all provided into the platform. But let's say I have a question. I'm making braised ribs, and let me tell you, moisture control is critical if I'm going to make that recipe. Now, if I ask a question like that, the agent is unable to answer the question. Why is that? It's because we haven't trained it in the context of how the La Cruset Dutch Oven product works. Now let's go take a look at what that product manual actually looks like. Look at this. That is incredible. A lot of time has been spent on building this content. It's got pictures, describes the product, ties the product's capabilities, different things that the product. And this is relevant, by the way, for that question that we just asked. But if we did this the old way and we took this content and just put it into a text document, it's going to look something like that. And that is just gobbledygoo. No one is going to understand, human or agent, how to use this information. Great. So now Allison's going to show us what this looks like if we take that piece of content and put it in our builder. So what we're going to do now is agentforce is going to take this very, very rich information and it is going to ingest it, it's going to index it. There's two steps. Number one, we're going to use AI to analyze this document. We're going to parse it, understand the pictures, understand where the words are, understand what connects to what, and come up with a recommendation for how we should index this data. Second, AgentForce is then going to apply that policy to the document. And this same policy can be applied to any document in the pipeline to easily scale it. Now we can test it right in the builder. Let's ask that same question and we'll see that we get a much richer answer and we get every reference that the agent has used to find this information. So we can check, was that the right reference, was that the right chunk where that information exists, or do we need to adjust this configuration, incredibly powerful, with a lot of detail to allow you to manage this context.

Great.

Now we feel good about it. We're going to publish, and then we're going to go back and see what this looks like. On William Sonova's website, we asked the same question again. And now with the power of intelligent context, we're going to get not just an accurate answer, but a rich answer that takes advantage of all this incredible content. And that is how William Sonoma makes olive cook. All right, to tell us more about how to build incredible experiences, the one and only Sanjna.

Here at the agentforce keynote, you get product and puns. Well, we're talking about what we've been learning from our customers over the last year of AgentForce. And one of the most critical things that we've absorbed from all of you is that you're trusting AgentForce with your most important asset, and that's the customer experience. And this experience represents your brand. And you need a way to make sure that that's as engaging and impactful as the experiences that you already have today. Because the fact of the matter is, is that your customers expect a unified experience no matter which channel they're using, whether they come into a store, they go into your app, or they pick up the phone. Now, the reality is that if you call a business today, the customer experience kind of starts to feel like your org chart, right? You got your sales department that has built out agents to help with sales. You have your service department that has an agent to help with service. You got your billing department that has their billing agent. You have all of these agents in isolation, and you don't have that unified experience with the same seamless handoff that every customer expects. Now, it's this learning that we've used to inspire with so much of our latest release, because we want you to be able to deliver a unified experience across every single channel. And that's why with the power of AgentForce Voice, you're going to be able to do just that. In this newest release, we're not only making voice available with the ability to create rich experiences like the ones that Madhav just showed. Now, why don't I show this to you in a live demo that kind of picks up where we left off with that Williams Sonoma experience. So you saw how we can bring all of this rich context into the experience. These images here that scroll from left to right, they're super engaging, just like I'm inside of the store. But Williams Sonoma also Wanted to bring the power of this experience to their 800 number. And soon you will all be able to configure voice right inside this agentforce builder that Angela and Aaron just showed you. You'll be able to configure the voice and the tone and give it a personality. Make that 800 number as engaging as any trip to your store. So why don't we see this in action by giving it a spin ourselves? What do you think, Alison?

Hi, welcome to Williams Sonoma. How can I help you?

Hi there. I'm ready to purchase the pan you recommended.

Hello, Sanj. I see you're referring to the copper saucier recommendation of a set of two for the Italian grape short route.

Yep, that's the one. That's the one. If I ordered it today, how soon could I get it delivered?

No problem. Let me pull that up. Based on your location in San Francisco, the scheduled delivery date would be Tuesday of next week.

That's too late. I need it for a dinner party on Saturday.

Okay, makes sense. I can create the order now and then transfer you to a specialist to review shipping options and in store inventory in real time. How does that sound?

Yes, please transfer me.

Thank you for your patience. Kindly hold as we connect you to.

The next available agent.

Connecting you to a Willem Sonoma representative. Great job, Olive. That was Olive speaking right to us. And that was our live Asian Force voice experience. I want to take a pause there to call out a few things. First is that Olive knew my nickname. How cool is that? Sanj, that's me. And I promise I'm not that rude in real life. But I wanted to make sure I interrupted the agent as it was so you could get a feel for how this technology works. Because we want it to be interruptible and to be very human, like in that way. So we saw that in action and behind the scenes. The story doesn't stop here because when I'm transferred to a live agent, they have the full transcript of that call and they can fully pick up the experience where we left off. This is an incredible way for you to create that unified voice experience that all of your customers are expecting. Now, we have a very special guest in the audience today. We have Samir Hasan, who is the CTO at Williams Sonoma. So, Sameer, why don't you come on. Come on up here? All right. Williams Sonoma, incredible trailblazer, Agent Blazer live with all of the AI sous chef agent in 28 days. Tell us how it's going and how you built it.

Oh, well, it's been phenomenal experience.

You know, if you know the William Sonoma brand, you know, we're about world class service. And even if you go back decades.

With our founder Chuck, Chuck Williams and.

His shop up in Sonoma, California, he would talk about the shopkeeper mentality. What does that mean? If you came into the shop, he would get to know you, he'd learn that you have kids at home, maybe you're a novice chef, you're hosting your.

First Thanksgiving and you're very nervous about it.

And he didn't just sell to you. He would set aside product, he would teach you, he would inspire you. And that's really built and infused into the DNA of the brand, into the.

Ethos of the brand.

And now with Agent Force and with Olive, we're taking that same shopkeeper mentality and we're bringing it into our digital channels.

I love that ethos of the brand. Just showing up through the agent. That's exactly, I think what everyone here is looking to do. Now, Williams Sonoma has eight different brands. You guys are deploying Agent Force all over. What are some of the learnings that you've had in being kind of an early adopter of this technology?

Yeah, we've learned so much working with.

Salesforce and with this amazing team over here. And you mentioned it, all of Soup.

To nuts 28 days. How did we do it? You know, part of it is these rock stores over here and a great.

Collaboration with the Salesforce team. But a big part of it is.

We took an AI first mentality and approach, not approaching building AI like traditional technology.

What does that mean?

We built agents that would comb over years of customer service. Transcripts, internal knowledge, articles, video content, unstructured content. And we use these agents to actually categorize and even help us build the agents themselves. We built other agents with different Personas to test the experience. We built a third set of agents that would then score those assessments.

So it's like agents on agents on.

Agents on the agency.

Yeah, a whole crew of agents. I love it.

Yeah.

And really that's why, why we're able to move so fast. So, you know, great execution, great partnership and really approaching it in a different way.

That's an incredible story. Let's give Sameer a hand. Thank you so much.

Thank you.

All right, now in our next chapter, we're going to learn how to turn all of you into agent managers and to bring us home. I want to welcome the one and only Leah McGowan here.

Thank you, Sanj. All right, that was Pretty awesome. And I'm super excited about our technology and the evolution of it. We saw how to build a controllable agent check. We saw how to provide IT context check. We created an amazing rich experience check. And we've deployed the agent. So we're done, right? Yeah, everyone, we're right.

You don't know.

Y' all just don't know.

Okay, we're not, we are not done. Everyone wants to know if their agents are performing well and if they're delivering consistent outcomes. So yes, everybody. And when, I mean say everyone, I mean everyone. We're talking the CEO levels, we're talking line of business leaders, we're talking the developers, your admin. Everyone wants to know how to manage their agents and scale their work. Because in this new era of agentic enterprise, we're ushering in an entirely new set of agent managers who have teams of agents working for them, focused on their business outcomes. And these agent managers, they're going to need tools. They're going to need tools to watch their agents, to make sure they're measuring and monitoring and optimizing all of the performance of the agents that they have deployed. Well, we have created some amazing tools to do just that. We're making agent managers lives easier by giving them visibility into the performance of their agents. Introducing agent Force observability. It is a single control center where agent managers can go in and optimize the performance of an agent. They can get alerts if the agents go a little off task and they can fine tune their behaviors. So that's great for managing agents, but what about scaling? Well, listen, I know at your organizations you got some folks building agents, you got some folks building workflows, you have other folks building prompts, you have all of these pockets of productivity. But we want to unleash the power of AI for everyone. We want to unleash it for your teams, for the entire enterprise. But in order to do that, you need tools that are going to leverage all this amazing work you've already done. All the work you've done in building AI agents, prompts and workflows, you need tools that are going to build end to end AI workflows in batch, turning those agentic processes into scalable business outcomes. And that tool, we got you covered. That tool, my friends, is AgentForce Grid. It is the easiest way to design and prototype AI native workflows and run AI at scale. I mean, in the ui, it can't get any easier than this. It's a spreadsheet. Everyone can run agentic Processes and get the feedback and the insights in minutes, in real time and to show us how to run and manage our agents. I am honored to bring to the stage the hardest man, working man in the business. Come on up here John, and show us how to manage our agents.

All right, thank you, Leah. All right, so we're going to go back to one of our mission critical agents that we have for AgentForce, the AgentForce help agent. So this is the platform that we have for agentforce to build not just any old agent, but the enterprise grade mission critical ones. We've done 1.8 million conversations and through that journey we've seen we need some new tools and that's what caused us to create Agent Force observability. Here. We're looking at Agent analytics powered by Tableau next, which gives all you business leaders the insights you need to see all your KPIs. When we do different experiments to expand this to different languages, to do new tests on the instructions we'd see how did the escalation rate change, how did the deflection rate change? Are we still delivering the CSAT that we need to, to make a great experience for all of you? But it's not just the analytics. We wanted to go one step further, which is why we added Agent Insights. So with Agent Insights, this is a performance review for our agent. Every session that gets created gets scored and we aggregate that all up so we can see how are we doing on creating cases, how are we doing on authentication. And we can see here our password reset is not exactly meeting expectations. So what we're going to do is dive in to see under the covers what's going on. So with our moments we have a list of every single conversation that you all have submitted to the AgentForce help agent. With this we can go another step deeper so that all the operations managers and all the agent managers can see what's going on. So we click into one of those moments and we can see the whole transcript. We can see the quality score for this one is low and we can very quickly identify the root cause. It looks like we're having some problems with single sign on timing out here. And so what we need to do is then see is this an isolated incident or is this something that's repeated? So what we're going to do is go through every one of those 1400 different sessions. Wait, no, we're not going to do that. We don't have time for that. We're going to export all these to AgentForce Grid. Let's open these in AgentForce Grid and see how we can analyze and optimize with anybody at scale. This is Agent Force Grid, our new AI workflow tool. We brought in all those moments, like any great AI tool, we're going to add in a prompt. For this prompt we want to figure out how many of these password reset problems are single sign on, how many of them are identity confirmation, how many of them are different issues. So we make that prompt, we link it to that first column, we hit generate, and this is where the magic happens. Unlike those other AI workflow tools, all of these execute in parallel. We're seeing every single row pop in. And this lets you iterate at the speed of thought, but it can do more than AI prompts. We're going to get a lot of reuse out of the other agents that we already use to support our human case reps. So we're going to add that agent in here. This case rep agent is going to do two different things. It's going to see are there known issues for these types of single sign on issues and we don't want to leave you hanging with a bad experience with the agent. So we're going to also create a case for all those sessions. So then we link it to that previous column here. By just adding the resource, we put that request to the agent and this agent is going to execute for all those sessions simultaneously to iterate on this workflow. When we get the output from that agent, we can pull out all the different details. So here we have the case information, we can pull out the case, we can click it with a single button. We just added another LinkedIn column. We keep extending this workflow kind of like automatic formulas in a spreadsheet. And what we want to do, finally, we want to update this workflow so that we can have a case that has the known issue. The support rep can get back to you fast with relevant info to have a great experience where our agent missed the mark. We're going to save this, we're going to run this. And this is the power of AI at scale for anybody in your company.

Super powerful. All right, you can clap.

All right, so now we're going to bring up the creator of agentforce Grid, Avi, to tell us about a bunch of other use cases.

Let's see it, Avi. Thank you, John. So when we started building agentforce Grid, we had one goal in mind. We wanted to make it easy not just for your admins, but for anyone in your organizations to build and run AI workflows. It didn't matter if you're in sales trying to research accounts at scale, in service trying to QA conversation transcripts, or in marketing trying to enrich and prioritize leads. Grid can do it all. John just showed a great use case of how to use Grid for a business problem like password reset. I'm going to switch it up. Let's do something a little fun. So yesterday, a bunch of you opted in to participate in today's keynote. I want you to reach into your pockets or your bags and take out your phones. We're going to use Grid to send you all a text message with a unique code that you can use to redeem a special agentforce T shirt after today's session. Sound good? All right, let's dive in. So here I am in Grid, and I've got all of your contact information loaded in this sheet. Each row is a contact record. Now, we talked to our lawyers. They said probably not a great idea to expose everyone's personal information on the big screen. So we've masked your data with a random number. But don't worry, at Salesforce, you're always more than just a number to us. So the first thing I need to do is I need to generate a unique code for everyone in this sheet. And I could use AI to do that, but that sounds a little overkill, a little expensive. So instead I'm just going to use a formula which are also native to Grid. And using Salesforce expression language, I can quickly author a formula that will output a random five digit number for each row in the sheet. And in real time, those codes will generate.

Great.

So I've got my contacts and I've got my codes. All I need to do now is send a text message. And what I can do is I can go ahead and use an agent. Now, for today's demo, I've built a custom Dreamforce agent which is capable of doing things like sending texts and answering FAQs. I'm just going to instruct this agent to send a text. And the cool thing about Grid is I can now reference data from the sheet. So I'm going to click Add Resource. Just select Contact. Great. I've got my contact information in here. And now I want to instruct everyone to pick up their T shirt using their code.

Great.

I'm going to grab the code. Okay. Awesome. So what I just did is really powerful. It's one instruction with dynamic context. So when this agent runs once per row or a thousand times, it's going to have the context that it Needs to send the right text to the right contact with the right code. All right, let's run it. So that's going to start to run over the next minute or two. You'll all probably start to receive a text message. Go ahead and remember that this is an agent. Feel free to ask it questions, have a conversation, have fun with it. Okay. And that is how you can run AI at scale with grid. Over to you, Adam.

All right, thank you, Avi.

I just got my text message. Going to pick up that T shirt.

All right.

We just saw a lot of stuff. But it boils down to what you need to know and the tools that you need to have to build great agents that work, the right tools to control them with precision. Being able to understand your data in the right context and bring it in to create ultimately great experiences that are rich and connect your customers where they are with things like Agent Force voice and all the tools to manage them, get the observability and drill down and then be able to take action. Like John just showed you about that password reset all together. AgentForce 360 is incredibly powerful. Now here's the thing. You can get hands on with this. Now these aren't features that are coming some other time. You can build your own voice agent right there in the campground to the right. Go ahead and do that. You get a phone number, you can take it home, give it to your boss and your colleagues. Experience Agent Force Voice now. And if you want to do the new builder and learn about agent script, you can do that in Moscone west on the second floor and get hands on. And we have our genius bar there of our four deployed engineers. And if you want to hear more about our customer success, go check out agentforce City and hear directly from them all while you're here. After the keynote and when you're flying home, if you're watching online, we got something for you as well. Join the community agent blazers. 93,000 and growing tons of amazing content community on slack and lots of great content on the trailheads. Become a champion. But most importantly, thank you and we cannot wait to see what you build. Have a great dreamforce, everybody.