Speaker 0: Hi, everyone. Uh, my name is Amber Juria Amel. I’m with Sercante. And today, I’m really excited to present, uh, or to introduce Josh Burke from Salesforce. He’s going to show everyone how to build your first agent in Agent Force. So take it away, Josh.

Speaker 1: Alrighty. Can you see my slides? Perfect. Alrighty. Uh, hello, everybody, and, uh, really a pleasure to be here, a pleasure to share, um, Agent Force with you. My name is Josh Burke. I’m an admin evangelist over here at at Salesforce. I’ve been with the company for just under fifteen years now. I’ve spent all of that time in evangelism, um, in doing new developer admin relations. And, um, it’s just it’s just kind of a joy. It’s a great little job where I get to run around and tell people what’s great about the technology, help people get enabled with the technology, um, and help you be successful with it. So what I’m going to do is I am going to go through a bit of an introduction as to kinda how we got here and the kind of technology that we’re talking about. I’m gonna show a quick demonstration, a little bit of a roadshow, shall you say, of some of the working technology behind, uh, building out an agent. And then we’re gonna go over to our good friend Trailhead, and we’re gonna get, um, hands on with it. So, uh, of course, it wouldn’t be a Salesforce presentation without our forward looking statements. Uh, please make all purchasing decisions based on current features and not any future looking statements I might make. Uh, thankfully, Agent Force is GA these days, and so there’s probably not a lot of future looking statements that I’ll be making.

Now just kind of as a quick history, like, why why did we why are we all talking about AI so much right now? Like, why is it so first and foremost, into our lives? And, effectively, what it comes down to is, um, it is the the size and scale of the recent AIs that have been have been developed. So GPT itself was one hundred and seventeen million parameters. Um, and at that time, none of nobody really heard about chat g p t one because, well, it wasn’t very good. Um, it wasn’t very smart. It didn’t behave very much like a human. It couldn’t do a lot. And so then they pulled up to 1,500,000,000, and 1,500,000,000 started to show some real interest, some real actual strength to it. But then what we saw, what got released to the public in the form of Chatubit is 175,000,000,000 parameters in size. And when we say parameters, those are all of the nodes within the model itself, within the large language model itself, that it is going to use to determine the question that you’re asking and try to generate the answer that you are looking for. And so the more of those parameters it has, the better it is at reading what you’re trying to say and the better it is at responding to it. Now to put that to scale, if Chatchip t two was the size of the moon, then Chatchip t three would be the size of the Earth. Like, that is how much larger of a paradigm Chatchip t three came to when it came to the amount of data that it has to pull. And to illustrate that in a different way, um, just how big because it’s kinda when we shift from million to billion, it’s kinda hard for us to sometimes think about, like, just how much larger is that. So if a million seconds is twelve days, then how long in time is one billion seconds? And the answer is thirty one years, which makes Chatibiti three five thousand four hundred and twenty five years in size, you know, compared to just seconds before. So these things, they’re big. They’re really, really big. And they’re basically you know, they’re scraping large portions of the Internet. Now, interestingly, what we’re seeing now is a trend not trying to push the size of these models larger and larger and larger. They’re actually trying to make them more performant. In a lot of ways, they’re actually trying to make them smaller. And one of the reasons why they’re trying to make them smaller is because they wanna put them on your laptops, and they wanna put them on their phones. And so this technology is becoming, um, it’s becoming more scalable, uh, and it is also something that is going to become, uh, even even more, uh, persuasive out there. It’s gonna be more and more places.

So how does a generative AI work? Well, we start with a prompt. And when we say a prompt, that is basically just the input that you were going to be giving into the model. Right? So very similar to if you go to Google, you put in a search phrase a search phrase, Google takes that. It goes through its rows and its columns, and then it’s going to give you the result. A prompt is something that you then give to the large language model, like write me a poem about Salesforce. And then it takes all of those parameters, and it looks at write me a poem about Salesforce, and it actually takes that apart. It takes right and me and a poem in Salesforce, and it looks at those, and it looks at it weighs weighs them. It’s like, what is important about this? Okay. I know you want a poem. I know I’m gonna need to know things about Salesforce. I know that I’m gonna have to connect these things together. And so once it looks at that and it realizes what it is that you’re trying to get out of the large language model, then it can give you based on previously existing data your response, your poem. In clouds above, Salesforce source, linking data, opening doors, lightning fast trails at pace, guiding businesses on digital ways. Now once again, though, this is a generative response. Okay? It is not the same kind of thing that we would get from a database. If you go to a database, you put in the same query and your data hasn’t changed, you’re always gonna get the same response. In a generative AI, if we ran that again, we might get a slightly different poem. It might do a slightly different style of poem. Um, and you could shift that prompt to give it more context to to lead it in that direction. Maybe instead of poem, you say haiku. And so when it goes into its parameters and says, oh, you want a haiku, a haiku has a certain number of rules to it, has a certain ways of thinking about poetry. We’re gonna obey those rules when we give our response.

And some basic terminology, um, and I really like this slide because it really does kind of it’s a really nice little tour of most of the things that you’re gonna hear about when it comes to generative AI. Generative AI itself, once again, is artificial intelligence that can that takes context, it takes a prompt, and then it can generate something. And that’s something, in our case, is mostly going to be text. It’s gonna be text based on Salesforce data, but it could also be images. It could be music. It could be video. To a generative AI, it doesn’t really matter. It’s taking in parameters and it’s putting out a response. One of the interesting things about GenAI is is they’re actually very good at things like translation because the difference to English, to German to it is really just a different set of parameters. And so as long as it has data about German versus English, it can take that, understand the context of one language, and then very easily put it into the other one. Uh, you’ll hear it referred to as the model, the LLM, the large language model, etcetera. The large language model is just basically that neural network of parameters for which it is pulling in, uh, the information to understand your prompt and give you your, uh, your response. The prompt, once again, is the starting line for a conversation. Now, one of the important things about working with agents and working with AIs is to remember that we’re in a conversational UI. So when we say a starting line, that’s one part of the prompt. And then most most of the AAs out there will allow you to then continue the conversation with the context that you have before. Um, and hopefully, we’re gonna see a little bit of that in a demo coming up here pretty quickly.

Now grounding. Context is king when it comes to talking to an AI. The more context that you can give it, the more accuracy you’re gonna get, uh, the more likely that you are going to get the exact kind of response. Uh, it’s going to behave itself in the way that you’ve described it because you have given it more and more context. And context can be rules, it can be roles, you can be telling some telling the AI, hey, I want to be able to, you know, I want you to to assume the role of a marketing user, and I want you to use only language that would be used in a professional context, etcetera, etcetera, etcetera. Now one thing I wanna stop right here, and this is something in my AI talks I talk about quite a lot, is we talk a lot about prompts, and we talk a lot about grounding. But I have a personal pet peeve when it comes to the phrase prompt engineering. Um, I think the phrase prompt engineering sounds like you need to get some union schmuck to come over to your house and rewire your laptop in order to get up and running with talking to your a AI models. There’s nowhere near the level of research and science required. Prompt writing is just that. It’s just writing. It’s having a conversation with an AI, and the best way to get better at it is just to do it. It’s just to just like writing itself, get in there, have the conversations with the AI, see what works, see what doesn’t, learn the little tricks of, like, oh, you did that because I didn’t give you a role. Right? But it’s not rocket science, and you don’t need, like, a master’s degree in order to work with it.

Now hallucinating and reasoning actually go hand in hand. An important thing to realize about AIs is that they are they are there to give you a human like response, which to the AI is going to successfully be what you were requesting in your prompt. And you might notice that I’m kind of dancing around by use of terms there because what I’m trying to do is avoid using the things like saying the AI is trying to give you the right answer. It is trying to give you the right answer. It would it will gladly give you the right answer if it can. However, accuracy and ethics are not its first goal. Its first goal is to give you a human like response that looks like what you wanted based on your product. And so if you don’t give it enough context, for instance, or it doesn’t have enough data to fill in the gaps, it will basically come up with English. It’ll come up with language that will fill in those gaps. And sometimes hallucinating is very kind of innocent in the sense that it is going to just sort of you know, your responses might change. However, hallucinating can get people into real trouble. And my cautionary tale that I always use for this is the lawyer who used ChatchiPT to create a legal brief, and he submitted the legal brief to the judge, and one of the judge’s aids realized that, uh, none of his precedents were real. ChatGPT had simply made them all up because it was like, you know, think CLIP. It’s like, oh, it looks like you’re writing a legal brief. I know legal briefs need precedents. I don’t really know any precedents for this legal brief, but I’m gonna make some that sound like they’re gonna support your case, and I’m gonna finish your legal brief with that. And this is why we say we still need the human in the loop. Right? We need we need to be able to be watchful in what our agents and what our AIs can do. Um, and we’ll see with agent builder, if you’re building an autonomous agent that’s gonna be exposed to, say, your customers, your job is to train that agent, test that agent, and make sure that agent is behaving correctly. Right? And so and then reasoning is is the, you know, the logical side of that. It is the AI putting all of this stuff, um, together.

So this is what I like to call my trust slide. Trust is our number one value here at Salesforce. Um, and when it comes to our Einstein platform and Agent Force, um, if your question is, can I trust this? Yes. Uh, this slide is very specifically designed to not have fancy animations or characters or anything like that. The answer is simply yes. Now you can also follow-up with that, well, how? And it’s a good question. And that’s because we have the Einstein trust layer. We are building an AI platform that runs on top of your enterprise data. It runs on top of, uh, your your enterprise working with your customers, working with your users. And we wanna make sure that we are keeping all of the privacy, uh, all of your data. We are not trying to to give you know, we’re we’re not making money off of your data. Right? And so we wanna make sure that when we are partnering with people like OpenAI or Azure that we are not sharing with them any any sensitive data that you have. And so that starts with building out that prompt, and then we do secure data retrieval within our platform within before anything gets sent over. We dynamically ground it. So we actually add in grounding rules to kinda make sure that there are additional guardrails to it. And then we put in data masking and prompt defense to make sure that anything that’s within your enterprise data is not actually being released out to any external model. Now this is really important because, for instance, OpenAI does make money off of your data. And this is one of the reasons I keep telling people, you need to start thinking about if you’re not using our tools, right, if you are using a tool like ChatJBT, then you have to remember everything that you send to ChatJBT, OpenAI can can see. As a couple of Samsung employees found out when they were putting in proprietary code into OpenAI, which meant that now OpenAI had access to Samsung proprietary code. Right? So we strip all of that out, and we give them a generic prompt that doesn’t show anything that’s going to be sensitive or any part of your your data. Um, we host as much within our model. We also will then use external models, and there is zero retention with that. We are we are if there’s anything that is getting audited, it’s being audited on our platform. And then the response comes back and the response comes back, and one of the things that we do is toxicity detection. And toxicity detection is a very important thing, and it’s basically it gets it gets scaled for what is being pulled into the response. And it is something that is monitored by humans, uh, over time. And the reason it has to be monitored by humans is because, once again, AI is not there to be ethical. AI doesn’t really have an understanding of what ethics is. It’s a language engine. And my cautionary tale here is always, uh, Microsoft before the Chat GPT craze. This is early days of, um, this it was actually closer to natural language processing, uh, which is sort of this when I was describing how you can take a prompt and pull it pull it apart, see its working parts, and put it back together again. That’s part of natural language language processing. And Microsoft had a, uh, a bot called Tay, and they trained Tay on Twitter and Reddit. And every time I do this live in front of an audience, like, the like, everybody’s face just, you know, kinda goes pale white because, yeah, sure enough, Tay turned into a racist hate machine within a couple of weeks and had to be pulled. Because Tay what Tay was reading was what Tay thought was ethical. And so if Tay sees a lot of unethical stuff, Tay doesn’t know that it’s being toxic. And so we still have this role. And in this case, I’m saying we, as in the Royal Salesforce, we have this role to make sure that it’s not growing more toxic over time. Then we take your data, we re we demask it. We put the enterprise data back in, um, and then we add it into the audit trail, which is part of data cloud, and then that then you actually see the results. And so all of this is to say that you can safely use things like prompt builder, which allows you to build templates in order to create AI generated responses over and over and over again on your records without having to worry about other companies seeing it, other companies having access to your data.

And we’ve gone a long way here. Right? It has been we are not new here at Salesforce to AI. Um, our early days included predictive analytics. My apologies. Analytics. Um, and, of course, we had Einstein bots. We had our original chat bots. And then things really started moving. We bought a company called MetaMind. And what MetaMind did was it was capable of understanding images. You gave it a large model set of images, and you could ask it things like, is this a fruit? Is this an apple? For one of our Dreamforces, I put together an application that tried to determine whether or not it was a real bear, a person in a bear suit, or just a person. Not exactly your most enterprise solution, but kind of a fun way to determine, like, you know, how an AI can actually determine what’s in an image. And then we added on Einstein Voice, which once again, that’s natural language processing. Anytime you talk to an Alexa or you talk to a Siri, it’s doing natural language processing. It’s determining what you’re trying to ask it based on what how your sentence is formed. And now finally, we’re in the world of generative AI with Einstein and AgentFlex.

Now you might have heard, um, the term Copilot. Copilot was, uh, the original name of the agents that we were using internally. This was before we started pushing our agents out to do things like customer bots and service bots and support bots. And Copilot basically has the same things. It’s it’s a LLM driven. It builds responses from actions, and it’s it’s it’s omnipresent within the internal Salesforce UI. Right? So if you’re on a contact page, you can pull up Einstein. If you’re on a on a opportunity page, you can pull up that exact same agent, uh, from the interface. Agent Force, basically the same thing. What we did is we added topics and instructions. We’re gonna talk a little bit more about that stuff. Our topics and instructions to give it more card rails and to give it more context as to what it should and should not do, helping it be an external autonomous agent that can talk directly to to to customers. They are both using the same underlying engine. So if you hear Copilot, you hear agent force, they’re very effectively the same thing. You will still see Copilot from time to time within the internal system, and that’s that’s the reason I still have this slide is to kind of clarify any of this confusion.

Now the five main attributes of of an agent. We determining what the role is. And so the role is going to say, hey. Um, this is this is your job. Uh, the data, this is the amount of knowledge that they’re gonna be able to access. And then the actions, the actions is very much the core of this is what you your these are the working blocks that you can use in order to generate a response. And then guardrails, what they should and shouldn’t do, those are the instructions. And then finally, a channel. Are we doing this internally? Are we putting this on Experience Cloud, etcetera, all building on top of trust. Topics are basically groups of things that then we can put in both the instructions and then the actions. And I’ll I’m I’m gonna kind of skip ahead here so that we can actually see that live.

So I’m gonna change my tabs out here really quickly. I’ll stop sharing. Screen. And let’s start in agent builder. Okay. Once you get up and running in Trailhead, this is what you’re going to see within, uh, agent builder itself. And let me kind of walk you through the actual, um, here. I’m gonna pull this back real quick so we start at the beginning. Pull up our service agent. Go to open in builder. Okay. So what our topics do is our topics allow us to organize, uh, different things that we wanna be able to talk about. So so if we wanna talk about different orders and then, like, class recommendations, we wanna work with reservations, etcetera, we can we can define all of those within one topic. And then if I go into my reservation topic, you can see that this is where we start defining things like the label, the classification, the scope, what is your job here, uh, and then a series of instructions about what to do when it when it is occurring things. And it can be very specific it can be things like acknowledge the and validate the user, like use empathy, use professionalism, all the way down to very specific things like, hey. If I’m trying to find multiple bookings, use this particular action.

Now if we go into our actions, here we have a series of actions. The great thing is that we can actually have actions that are based on either standard actions or custom actions. And standard actions are going to be things like here, let me go over to my other topic. One of the first ones we used. Right. Okay. So these are actually almost all I think these actually are all standard actions. And it is things like querying the records and identifying the records by name and then creating a to do. Right? And so if I go over here, this is our conversation preview. So now this is gonna be the same exact agent that we’re going to see, um, internally or externally. And this allows us to test what it’s gonna be. So I’m gonna say something like find correct James. Now remember, I haven’t built anything custom here yet. Okay. Now I found a contact named Craig James. How can I assist you further with this contact? Summarize Craig James. And then I’ll walk you through what we’re seeing here. Yep. We got an error. Model uh-oh. My demos may be in trouble here. I think we have multiple people working off of them at the same time. Uh, anyway, so here is what they call the planner. I like to call it kind of the brain of the operation. It shows you exactly, uh, what it is that we are doing here. Right? So it took that prompt, and it it’s like, oh, you’re trying to find a record. I have the standard action and it identify record by name. I’m gonna take that, and then I’m going to go use that action to determine the information. And here we have, uh, all of the things that are going on. Now this is very useful if your agent isn’t working correctly. Right? Like, right here, when we get down to this error, I can actually see what the error is actually talking about. Uh, so that’s probably just too long of a of a of a response for what’s what we’re doing in this particular org. You shouldn’t run into this problem. But I’m gonna say let’s try this. Create a to do for correct James. Okay. Great. Uh, I follow, uh, up email due at the end of the week. And once again, so it has see, it already knows who Coretta is. Right? It doesn’t have to go back and find. It knows the context of what I’m talking about, and then it can go grab this standard action here, create a to do, and take that and then and then actually get that done. And so a lot of power out of agent force right out of the gate.

I wanna show you two more things, then we’re gonna hop over to the Trailhead. What I wanna show is First, this is a prompt. This is prompt builder. Prompt builder allows us to create one prompt that’s gonna be used over and over again based on the specific kind of account type. And this is one that comes comes out of the box, but it’s the if you’re creating a custom one, it would be pretty much the same thing. I really like this one though because it’s a great example of look how much context it’s being given. Right? How much context do we have here in order to see what’s going on? Now this can be used within Agent Force if you wanna say, describe an account in a very specific way. You could create a custom action that uses a custom summary and then instruct Agent Force to say, hey. When you’re talking about an action, try it this way. And so now if I search my accounts and preview, we see we’re getting HTML. Right? And as it was informed to do, and it’s pulling off of that account. And then you can also see what it was thinking in in the side over here. And so if you wanna customize how agent force is talking to a person, prop builder is one way to do it.

And then the other thing I wanna show is this. Now if you are already somebody who is even fairly familiar with flow, you don’t have to be the great Jennifer Lee, queen of flow in order to get agent force. So you’re trying to do what you want. This is a very straightforward flow, but it’s a I it’s an example of saying, hey. I wanna train agent force how to get a how to get a session based on somebody’s email, so like, correct email. And then I wanna find the sessions based on that. But then what I wanna actually give you back is the actual booking. Like like like so the class is, you know, English, but this is the class that was on Friday at 03:00. I wanna be able to make sure Agent Force, when it gets this question, understands how to do this. And within the trailhead, you’re what you’re gonna see is you’re gonna see that this is actually something. Um, you create the flow, very straightforward flow, and then you go into Agent Force. You create an action. You associate with the flow, and there you determine the inputs and the outputs. For the Apex developers in the room, you do not get to get away scot free. Uh, you can also use Apex with Agent Force. You can use Apex and Flow with prompt builder. For Apex, it’s the exact same thing that you’re used to. If you’ve ever built Apex, it’s gonna work with Flow. So it is, um, it it is that same, um, attributes in order to say, hey. I’m an a I’m an Apex class that you can, um, you could utilize.

Alright. Let’s get hands on. And then while we’re getting hands on, we can take questions. Let’s see here. Back to you. Share. Okay. So if you go to SFDC oops. I’m gonna leave this this up for a while, and I’m gonna flip over. Actually, view that tab. Okay. I’m gonna leave this up, and, um, I’m gonna start looking at the q and a tab now. Uh, and if you go to that sfc.co, you’re gonna go to Trailhead. You’re gonna need to log in to Trailhead. It is our Agent Force quick start, and you’re gonna get access to a very special kind of Trailhead playground. And that Trailhead playground is going to be already up and running, uh, with all of the things that you’re gonna need in order to get your hands on with Agent Force. You get if we’re you’re probably gonna go only get through most of it within the remaining fifteen minutes we have together here. Um, you do not necessarily have to start you have to finish it today, of course. It’s Trailhead. Feel free to to to take it home.

So give me a second to pause here and check the q and a. How to actually train the agent? And actually, question. So, um, you don’t have to train the agent in terms of utilizing, like, the model itself. Like, we handle that for you. Um, how to train the agent is mostly by either making sure it’s using so there’s there’s three main things. There’s the instructions, which those are the guardrails. Right? Those are the things that’s an additional level of context that’s going to be used every time it’s having an interaction with somebody. And so that is a nice umbrella to put in put in the things like be professional, be empathetic. Um, you can even I’m trying to remember exactly what I told it, but I you can even tell it things like if you cannot find the information. Um, actually, the one I did was remind the user that the information they’re looking for so remember we had that topic that was, like, reservation manager. Right? So I know we’re talking about reservations. Remind the user reservation info might be in their confirmation email. And so without adding a flow or prompt or custom action or anything like that, the agent started talking to the person like that. They’re like, oh, maybe you should check you know, it didn’t say quite. But, you know, please check your inbox to see if you have a confirmation email. And right there and, you know, we’re we’re seeing with correct where it’s you know, it understands the context. It remembers who correct is. It understands where the questions are going, etcetera etcetera. It’s it’s it’s it’s so much different. Right? As somebody who kinda hates talking to robots on the phone or online, like this new age of of generative AI agents that are friendlier, they’re more human like, they’re more responsive, It’s just it’s from a customer experience, it’s wonderful. So we have instructions. Uh, we have our standard actions, and you can either limit the standard actions or add in more to make it more powerful, and then you can add in custom actions. And as we are seeing with the flow, the custom actions help train the agent if you wanted to perform in very specific ways. And the other thing I wanna point out is that the flow we saw, that demonstration demonstrate flow that had, like, five to seven, you know, steps to it, that’s the standard we’re usually seeing. In fact, it’s the recommended one. Um, it’s better to have three or four small flows than to try to do one flow that does everything all at once. Because remember, what the what agent force is gonna do, it’s gonna go look at that library of agents and be like, all those those library of actions, be like, oh, which one of these will help me get this done? And so if it has three different ones to choose from, you can put two of those together and respond it. Or maybe sometimes it’s the other two, and it’ll put those together to respond. So that’s how, you know, the training really works. And the process really is is getting in there and trying it, trying different things, try different piece points of data, um, and then see how, you know, how does the agent respond. And then if it’s not responding correctly, start with the instructions because that’s that’s no code. Right? See if you can get it to work more properly using instructions. And if that doesn’t work, maybe a small flow. You know, maybe a short flow will help it get to the bottom line. Flows are really, really great when you don’t wanna rely on the agent from thinking too much to is is kind of one way to put it. Another demonstration I’ve done previously is, um, looking at all of the bookings somebody did and then telling me how much they’ve spent. Right? Because let’s say you’re you’re you’re in customer support. You wanna know, um, is Craig James, you know, somebody who you know, all of everybody’s important. Right? But but if Craig has spent $50,000 with us, maybe refunding that last workshop she did isn’t such a big deal. So what I told the agent to do was go through all the bookings and then summarize how much we’d spent. The problem is it would hallucinate it would get everything right, but then it would hallucinate the dollar amount amount because it’s not actually all that great at math. Once again, it’s a language model. So then I just built a simple, like, three step flow, pointed that flow to the prop builder, and everything worked correctly.

Okay. I’m gonna hop over to Trailhead itself, uh, because there is a couple things in I’m assuming everybody has that link, and I see that we have it in the chat. Give me a second to log in to uh, Trailhead itself. Okay. Um, just a couple things to point out. First of all, this the connection to the playground and I am getting a different message here that you will probably get. This is this is the one, um, I really wanna point out. Password resets may take a few minutes. If an error occurs, wait a minute and retry with the same credentials. There is also a scenario where you don’t get an error. It just tells you that your password is wrong even though your password is not wrong. Uh, so if you are having troubles logging in, you you set up your new password, you’re pretty positive, you have your new password, you are not crazy. It is just the fact that spinning up these orgs are a little bit more take a little bit more oomph than the usual Trailhead Playground. Um, and usually, that should resolve itself within about two to two to three minutes, maybe five minutes tops. Um, just go grab a coffee or check a Facebook feed or something like that, and then that should resolve itself. But it is it’s not it’s not something that you are doing, um, incorrectly. The other thing that is very important is to make sure that you do all of these early steps in the right order, um, and don’t skip any of them. If you this is fairly straightforward. However, for instance, if you start using like, there’s instances where Prop Builder will be there, but it won’t work correctly because Einstein’s not turned on correctly. Once again, you’ll get an error that’s like, this prompt does not work, but it’s not very descriptive that you should go you’ll make sure you toggled everything on correctly. Also, if you ever have trouble finding something in setup after after you’ve turned on anything, Einstein, just refresh the page because sometimes setup has changed because of the new things that have been enabled, but the browser has not seen it. So if you think Prop Builder should be there or agent should be there and you’re not seeing it, try refreshing the browser and then searching in the quick searching app. Uh, so for instance, this is the part that is if you don’t if you skip this and then move on to this, Experience Cloud will not work correctly till later on. Oh, yes. Um, in in case you happen to be, um, in in New York next week, uh, we are having a hackathon on November, uh, with a wonderful many wonderful prizes, the grand prize being, uh, $20,000. Um, it’ll be a great way to get hands on, uh, with Agent Force, with Data Cloud, with Prop Builder, all of these wonderful things. It is going to be at the Javits Center, November. And we might be sold out. I’m not sure. Uh, but it will be a great way to partner up with people, uh, hear from experts, um, get your hands on with it, have some fun, and maybe grab some cash. Thank you, Michelle. How setting up an agent for customer versus the employees is different. Um, so that is basically the experience of exposing it through experience cloud. And so that is the one thing that I didn’t really talk a lot about is when we refer to a channel. Um, and so the channel sort of defines, are we doing this for an internal user or for an external user? The nice thing though is that everything else we are talking about. Alright? We have we have our rules. We have our we have our training. We have our instructions. We have our actions. All that remains the same. So there’s not any one specific thing to do that’s that’s different other than realizing that, um, when you’re when you are inside the system, when you’re inside the platform, you are going to be you’re going to know who you’re talking to. Right? They’re they have a user. They’re they’re logged in. When it comes to an external, you’re you’re probably starting for more of a blank slate. So you have to consider that, uh, within your design.

Speaker 0: Thank you, Josh. Um, it looks like you have about a minute left. So if anybody has any questions, please, uh, post them in the chat now. But, um, if there are no more questions, then you can reach out to Josh directly. Um, you can find his speaker profile, um, in the speakers tab in the Marjorievan platform, um, and you can ask him any questions about the Trailhead or about building your first agent, anything like that.

Speaker 1: Yes. And please do so. Uh, it is my job to talk to the community about these things. So, you know, I have people like, oh, I’m sorry to bother you. I’m like, no. I’m actually getting paid for this. Uh, the the joke I usually tell is I am available for children’s parties as long as your child wants to hear about the Salesforce platform. So feel free to reach out to me on LinkedIn or Blue Sky, um, or on the various Slack. I’m usually on some of the Slack communities out there as well.

Speaker 0: Awesome. Well, thank you, everybody, and have fun at your next session. Bye.

Speaker 1: Thanks, everybody.