See What's New in Agentforce | World Tour NYC 2025 | Salesforce

Hello everyone and welcome. My name is Austin Javaliana and I'm a Product Marketing manager here at AgentForce. As a PMM on the AgentForce team, I work with all of our wonderful product teams to bring everything AgentForce to you. So AgentForce builder, AgentForce vibes, really anything with the name AgentForce on it, our team is likely involved Today. I'm so excited to be taking you through some of AgentForce's newest innovations. We're going to do a recap of what was announced at Dreamforce and then dive into a few innovations that you probably haven't seen yet. As always, if you have not done so already, please download the Salesforce mobile app. For those in the room would love your feedback on this session as well as the overall event. Your feedback is very important to us. It helps us improve our sessions for the next time that we see you at an event. And as always, a huge thank you for me. I want to thank each and every one of you for being the driving force behind behind agentforce. Wanna thank the 12,000 customers out there already building agents and helping transform customer and employee experiences. And I wanna thank you all. It's 3:30 late in the day. Thank you so much for joining us for this session. So let's dive in. Before we start looking at some of the innovations, let's dive in and review a little bit from that keynote. As we talked about, we're in this technology revolution and what we're seeing is we're witnessing AI be integrated into the daily lives of us faster than ever before. For me personally, I know it's now a rarity. If I'm not on ChatGPT on the way to work on the train, or if I'm snapping a picture of my food to help AI, let me know how healthy or right now, in this holiday season, unhealthy I'm being. But for the enterprise, that integration of AI into the business hasn't been as fast and it hasn't been as fast and as easy for a number of reasons. First is the tools that we use, they're disconnected. Having AI be working across disconnected tools just doesn't work. Second, those tools are fairly manual. And third, when we think about AI and the way that we use it, the way that we're implementing it for our employees and our customers sometimes doesn't meet our expectations at this point. And that's why over the past year we've been working side by side with our customers to help them bring AI and agents into into the flow of work across their Organizations. Over the course of four releases, we've built the tools, infrastructure and out of the box resources that you and your organizations need to get up and running with Agentic AI. And now with our latest release, Agentforce360, we're helping bring agents and humans together to connect with your customers, employees, operations and agents in a whole new way. So with Agent Force 360, we're really doing similarly what we did with CRM where we took CRM at Salesforce to connect not just to sales, but sales, to marketing, to commerce, to field service. And that's the way we're thinking about AgentForce360. We're turning all of our products agent first, putting data and AI into every app to help orchestrate those workflows across the org. With AgentForce 360, there are three important layers. First is the customer layer. This is the foundational app layer. This is the app layer you're probably all used to with sales, service, marketing, commerce. The second is our data layer, Data360. In Data360, this provides that unified real time view of your customer that helps you when all of your workers and employees and customers are actually using those apps. And then third is the Agentforce360 platform, which we're gonna talk a little about today. The AgentForce 360 platform. This is your Agentic layer. So these are the actual tools, resources out of the box resources, like I mentioned, that will help you transform those customer and employee experiences into agent first experiences. The AgentForce360 platform, this is the AI agent platform for every customer and employee experience. And we're gonna do just a quick recap of some of our features that were released at Dreamforce within this platform. And we're going to start with our builder. So AgentForce builder, if you've used our old builder, you're going to notice a humongous upgrade and a different experience. Instead of just typing in instructions and sending that to the LLM, you can now build your agent in a document like Canvas View with natural language. So what that means is in this document, Canvas View, you can actually type like you're typing to an LLM what you want your agent to do. But if you ever get stuck, you don't know where to start, or if you start building and you're starting to unit test and things just aren't looking right. Within the new building experience is an AI assistant builder right there with you. So whatever spot you are with building that agent, you can ask this assistant, hey, I'm getting A failure here. This agent's not responding the way I need to. What's your suggestion? And that AI assistant will actually make a suggestion and put that in the canvas for you to either accept or decline. And if there's one thing we heard about the building experience over the course of the year is that you need your agents to perform predictably every time. That's why within this new building experience, we're excited to introduce Agent Script. Agent Script is a new scripting language that allows you to build guardrails into your agent. So before we had script, what we were doing was we were writing instructions like always do this but never do this, also do this before this happens. And sending those instructions to the LLM really confused the agent and brought back unpredictable results. And as we all know, there are many business processes brand things that you need those agents to follow. So with AgentScript, we're taking that kind of unpredictability out and allowing you to build those guardrails in so the agent performs exactly how you need it to every time. Another AgentForce 360 platform announcement is AgentForce Voice. So we're extremely excited that AgentForce Voice is now in GA and we're excited about this because over 80% of service happens over the phone. So with AgentForce Voice, you can build those agents and have a like, human, like voice in your brand and tone that can help employ customers and employees over the phone. And we're very excited for this to be in GA now and what's coming up in the future as well is agentforce Voice within digital channels as well as additional language support. Now all of this is great, it's fantastic. We're super excited about this, but it doesn't really matter because if your data's wrong, that agent can't work properly. So the other innovation I want to share with you is intelligent context. With intelligent context, you can now take all of your unstructured, messy data, your PDFs, your images, those JPEGs that you want the agent to use and actually use AI to parse them out and help the agent understand actually what those things are so that they can provide accurate answers to your customers, employees. So now I've been talking quite a lot. Let's actually take a look at what intelligent Context looks like.

Answer to this question is inside this PDF, which contains a combination of visual graphs and formatted information. And if we scroll below, we'll see the answer is very clearly 520. So what's the problem? Well, we, when most agents index content like this for retrieval, it becomes something that looks a bit more like this, where all the rich context gets collapsed and it becomes impossible to understand what 520 actually means. And that is why when I send this question to most agents, it responds with the wrong answer. Because making agents accurate isn't just a data problem, it's also a context problem. So how can AgentForce help? Well, this is intelligent context, which allows you to fine tune your search indices while preserving that rich visual context. In this workspace, you can upload sample documents like that PDF you just saw, and create a tuned index for similar documents. And once uploaded, you can have AgentForce find the best indexing strategy automatically. This intelligently determines the best strategy for parsing, processing, chunking and embedding. But we can of course fine tune all aspects manually. And one of the big breakthroughs is the option to use LLM parsing to understand the context of the data instead of just the raw text. What does that really mean? Well, let's look at the extracted chunk that the agent previously struggled with. Notice that it now correctly summarizes the information about the flight hours, preserving the context. This context awareness also extends to images, charts and diagrams with LLM based image processing. What actually is context awareness? These LLM processing steps work with a prompt, and every good prompt starts with a Persona that the LLM should assume. So this ensures that the data is interpreted as your customers would too. And once you're finished testing your new index with few sample documents, you can apply it to a library of thousands of other documents which your Agent Force agents can tap into. Which is why with AgentForce we can ask this agent that same question. But now we'll not only get the right context aware answer, but also citations that provide transparency into how AgentForce came up with that answer. So this is how intelligent context gives AgentForce deeper contextual understanding of your data and the ability to provide better answers. To learn more, visit agentforce.com thanks for watching.

As you saw, with intelligent Context, this is really a big step forward to making sure those agents have all the data they need in a format that they can use again to increase that predictability of their results with your customers, employees. And finally, just to touch on this quick, another really great announcement. Agentforce Vibes with agentforce Vibes, you can build pretty much anything in salesforce, turning your ideas into apps, workflows and agents. When you build with Vibes, you're building with an AI assistant that understands your business. It's built on the same Salesforce metadata that the rest of your organization's built on. So when you're building with vibes, you're building with a partner that knows your business very well. Now that was Dreamforce. And if there's anything that I know about AgentForce and being on this team is that we move fast and that we're always working to improve your experiences with agents. So I'm excited to introduce you and show you a few other innovations that I think you'll find interesting. The first, we're going to talk about Agent Force Observability. What happens after you deploy your agents? We'll dive a little bit into connected agents and some MCP support coming up. And then finally a really exciting announcement. Agent force sales in ChatGPT starting with agent force observability. This just went GA a few weeks ago and what we hear from customers is building the agent, deploying the agent, hitting that launch button. It's a party. It's amazing. We get it out there, we're ready for customers and employees to experience it. But that's really when the work begins. Because once that agent's out there, we need to start asking ourselves questions. How's it doing? What's it doing? What's it actually responding? Like it's not enough just to build and deploy the agent. We need to make sure it's working and that we're optimizing it for our KPIs. And that's why we asked ourselves at Salesforce, how do we help you manage, measure and optimize your agents? And this is where AgentForce observability comes in. You can think of Agent Force Observability as your single mission control for unified visibility monitoring and optimization of your agents. This is accessed directly through AgentForce Studio and it's built on our session tracing data model. And what that means is that Agent Force Observability is is monitoring every interaction your agent is having all the way down to a single conversational turn. Really what we're trying to get here are more accurate results, faster resolution and better customer experiences after we deploy those agents. So now I just want to show you a quick demo video of Agent Force Observability.

We're inside AgentForce Studio, the home for deep observability for AgentForce. 360. I start with agent analytics and with the power of Data 360 I can observe my service voice, STR employee and ITSM agents all in one place. I instantly gain strategic oversight into usage adoption and a high level overview of all the other available agents. With Agent Effectiveness, I dive deeper into agent performance, I can track session outcome trends and analyze the granular topics and actions usage to measure efficiency and determine where the agent changes are needed to ensure the most robust performance. Now I also want to make sure that my agents are actually helpful. And that's why every agent is given a quality score to assess the agent's response to every customer request. This quality score is calculated using LLM as a judge. A quality score is judged not just across topics, but also across similar intents expressed by customers in thousands of conversations that are grouped in what we call intents. This way, the identified areas of improvement are the most precise. Now let's examine one of these intents. This is the bridge between seeing the trend and understanding the behavior. I can see the exact interactions within this intent, but I want to find out precisely how the agent was thinking in every response. Response. The entire span of the response, including the reasoning path and the output evaluation is in front of me. This gives me incredible power to see how the agent is thinking in real time. And that is because of our session tracing data model. I can drill into each step of this resolution path, spot any errors and identify the root cause for the a clear path to resolution. Collectively, the quality score, topics and intents are extremely powerful levers to understand exactly how the agents are performing. And they give us the tools to make these agents even better. And that is exactly how observability with AgentForce delivers the clarity that customers need to trust their agents and scale them across their enterprise. To learn more, visit agentforce.com thank you for watching.

So with that demo with AgentForce Observability, you observed a lot, but you also have to think there's going to be a lot. We're going to have a lot of conversations, a lot of back and forth with agents. And what agent Force Observability will do is help you really hone in and understand exactly what those agents are doing and and how to improve them. So to summarize observability, just a few things to think about and how it kind of breaks down. First is agent analytics and that's gonna help you understand how your agents across the board are performing. The second is agent optimization. And that's where you could really dive into a single agent, into a single conversation that that agent's having and understand, okay, how's this one performing? Agent analytics. Agent optimization is in GA today. So if you're building agents, please feel free to use that. And coming early next year is going to be agent health monitoring. And with agent health monitoring, that's going to be real time, near time, real time views of your agent. So we're thinking what's their uptime, what's that escalation rate in real time and what's latency that's happening. So a lot more to come with agent Force Observability. But we're really excited to get your hands on with agent force observability. So you can see how those agents are performing. The next innovation to talk through is connected agents. So when we think about connective agents, we've noticed kind of one or two problems our customers have. The first is they build a somewhat generic agent or one agent and they want that agent to do a lot of use cases. But what happens is when you give one agent a lot of use cases and if those use cases are pretty different, the agent struggles to perform them. It's just simply a little too much for that agent to do. On the flip side, we have our customers that build highly specialized agents. They're very good at a single use case. They perform well every time. But what happens there is it creates somewhat disjointed customer experience. You know, you're sending your employees to a product support agent, but then, okay, you need to go to the HR agent to do this and then go to another agent to do that to solve for this. What we're introducing and what's in pilot now. Well, what we did is what we asked if we could deploy a team of AI agents with the ease of a single agent. Which leads us to what's in pilot now, which is multi agent orchestration. So with multi agent orchestration, what you're really doing is building a single kind of front door agent to your brand. What that agent will do will interact with your customer, with your employee, and help decide which agent it should call in to actually support the customer, the use case, the employee, whatever it might be. What's great about multi agent orchestration is it's fairly low code to build and actually connect your agents. So you'll be building your agents individually and then this front door agent, you'll be able to build and connect all together. But multi agent orchestration is just part of a larger picture. So when we think about connective agents, we're really taking a similar approach to what Salesforce has always done. We're making our system open and extensible. So we want you to be able to connect your agents to other agents with integration in your organization, you know, within your boundary, as well as externally to other agents. So we have multi agent orchestration in pilot now upcoming is open agent collaboration. So using protocols like agent to agent, protocol A to A so that you can start connecting your agents to external agents and having them run all on your data. That's safe and secure in Salesforce. And lastly, another really cool announcement that just came out very recently is Agent Force Sales and chatgpt. So kind of building off that idea of connected agents. This is a really good example of how we're making Salesforce open extensible for the agents built with us. So with Agentforced sales and ChatGPT, what we'll be doing is bringing agent for sales into ChatGPT so that sellers can do a number of things. First is to give ChatGPT CRM context. That means your seller can be looking up information about a prospect and also ask questions about that prospect about the information that's in Salesforce. So they could ask about account details, customer insights, you know what's happened with the customer and have all that information directly in ChatGPT. Additionally, they can actually take action within ChatGPT. So if they need to move a stage of a deal, they need to update a record, they can do that directly in chat and that syncs back to Salesforce so that seller can really live their day, day to day life right out of ChatGPT within whatever selling process that they're in. And within those two things it should be noted that data integrity and how that data connects is there. So when they Update something in ChatGPT that's directly linked within Salesforce, if they update something in Salesforce and they start using ChatGPT to ask questions about it, that information is a two way street. So that seller is always working off the the correct information. And what's really exciting is I actually have a demo of this one. So let's take a look at this. Agent for sales in ChatGPT.

This is agent for sales intelligence delivered in ChatGPT. As a seller, I'm starting my day by checking in on my hot leads. Bringing Salesforce into ChatGPT allows me to gain insights and prioritize how I want to spend my time. I can use ChatGPT to do research and use this information to enrich my Salesforce leads to and take action in one place. One prompt pushes everything back into Salesforce. No forms to fill out, no fields to hunt for. Data hygiene happens automatically improving forecasting and pipeline accuracy. Now that I've qualified this lead, I'll hand it off to my AgentForce sales agent. The agent will complete outreach using the enriched Description generated by ChatGPT and saved to the Salesforce record. Fast forward and I'm out on the road. With a busy day of meetings Scheduled, I'll use ChatGPT Voice to ask what do I have on my calendar today? The ChatGPT integration gives me access to all of the power of Salesforce Hands free on the go. I remember meeting with Mark a month ago, but I can't remember what we talked about. Let me ask what did I talk about in the last meeting with Mark? With Salesforce data at my fingertips, I am prepared for the meeting in seconds. The meeting went really well and I want to send a quick follow up. Based on the meeting transcript stored in Salesforce, I can generate a highly personalized follow up in seconds. As I complete work in ChatGPT, I can easily keep Salesforce up to date by creating records. Now that this deal is really taking shape, I want to do some planning with ChatGPT. I'm creating an account plan. As we go from prospecting to closing, every moment is faster and easier. With Salesforce and ChatGPT, I can outline my pitch deck and objection handling will help me feel prepared heading into our big meeting. All of this preparation is made possible by the combined power of Salesforce and ChatGPT success. I won the deal. I'll update the opportunity in Salesforce and post a deal recap in Slack so I can celebrate with the team. For sellers, Salesforce in ChatGPT means higher quality outreach, lower faster follow up, up to date data so sellers can spend more time closing.

And as I mentioned this is really just the start with AgentForce. What I've noticed about being on this team is that we ship very quickly. So for the Agent force sales and ChatGPT that's tentatively moving into beta in December. So if you're interested please reach out to your account rep your AE to get connected. But we have a lot more coming. So for AgentForce, just a few highlights for you. As mentioned with Voice we're extending voice over more channels so we'll have support for session initiation protocol, digital channels and a lot more language support in the upcoming year. Agent Health Monitoring again that real time view of your agent. So latency uptime escalation rate that's slated to go later in the new year as well. A2 a third party orchestration as I mentioned again in the new year and one we didn't touch on in this presentation but I want to make sure you're aware of is AgentForce Grid. AgentForce Grid if you're not aware this is our spreadsheet like experience, where you can create AI workflows to batch test your agents, you can create AI workflows to enrich leads, you can create AI workflows to prototype a number of different things. Right now what we're doing with Grid is you can pull in records, you can pull in prompts, you can pull in your AI agents to string these workflows together. And in the future you'll also be able to pull in things like images, PDFs against more of that unstructured data to test and refine AI workflows that are using more data types like that. I know it's late in the day. I want to thank you again for coming. If you're sticking around, we have an amazing deep dive into Agent 4's voice. So if you're interested about Voice, how that's being deployed, that's going to be downstairs in theater one at 4:30pm for those of you in the room, and if you haven't done so already, we have a number of demos in campground. A lot of the things I just discussed you can go get hands on with today. And I just want to say thank you so much for coming and thanks for your time. And again, thank you for being that driving force behind AgentForce. It's been incredibly inspiring to see what you're doing with this technology. Thanks y'.

All.