Salesforce AGENTFORCE Demo Session 1

[00:00:06]Okay, cool. Uh, so let me get started.

[00:00:14]Dis can you just give an uh option to share my screen? I'll just send you the request.

[00:00:22]Thank you.

[00:00:27]Okay. So I hope you all know why we are here right. We are here to learn one of the most trending or topic in the market that is agent force. Before I start to talk about that particular topic does anyone know or anyone have any idea about it?

[00:00:48]Anything that you know about obviously we all are here to learn about it but just wanted to check the pulse of the audience. Anyone here know about it? It's okay if even if it is right or wrong. I know Monday mornings it's very hard to, you know, get up and then to unmute yourself and speak. But if you're willing to do so, please.

[00:01:20]Okay, I see someone putting up it in the chat. Salesforce support chatbot redirect. Okay, close to the correct one. Nice babu.

[00:01:29]Anyone else? Okay, let me turn off my camera. It should not seem like I'm looking at you and asking you continuously.

[00:01:37]Okay, so it seems like everyone is in the same boat the way I was there two years back. Okay, and everyone is just getting started. So that's with that said first you should know the instructor who is teaching you is eligible enough to train you on that particular topic or not right so you're going to learn agent force from me in next 30 days starting from today right so before I talk about agent force I should talk about me so right so my name is Amantiari and I am in this Salesforce ecosystem from last 8 years close to a decade and I've been working on agent force right from the time it was into very initial phase like if I tell you from when I'm working on agent force I'm working on it from the time it was into the face of co-pilot okay from that time I've been working into the agent force and I'm also a salesforce marketing champion and I have and I have trained more than 3,500 plus students in my 7 years span of training and apart from this I also write blogs I also of create YouTube channels and I also lead community group in Mumbai.

[00:02:51]Okay. So that is about me and I am a tech lead in a in a MNC. Okay. So that is about me and today we are going to start with a very simple thing like we will start with introduction to AI and agent force. I hope you all know what are we going to cover throughout this. I mean, do you all know, just give me a thumbs up if you know it or thumbs down if you don't know it. Do you all know what are we going to cover throughout this training? What are the different topics? What are the different uh things that we're going to cover in this agent force training? Quick thumbs up if you know it. Thumbs down if you don't know it.

[00:03:33]Okay, I see one thumbs up.

[00:03:38]others there is a reaction button which we which you can use. Yes, I can see three four five five thumbs up coming up.

[00:03:47]Okay, that means many people know about it what we are going to cover in this training. Right. So yes, I will be starting with basics right from the scratch. Okay, what is AI and then we will understand from how this AI agent force got introduced in Salesforce. Then we are going to see the different types of agent uh you know actions topics and everything in depth. Then we will see rag that is retrol augmentation generation which is a part of data cloud that we will see. After that we will see what are the different types of agents that are available. After that we will see what do we mean when we talk about model builders when we talk about bring your own models. Okay. After that we will see MCP. Now you might have heard a lot about Salesforce is promoting like anything headless 360 right how this headless 360 came into picture all that things we are going to see in depth so we are going to scratch from going to start from scratch and we are going to uh leave it or we are going to end it at the advanced level okay so with that said what that comes to your mind when you hear the term AI I will not talk to talk about agent force I will talk about AI give you two minutes to think about it but before that what are we going to cover today so first we'll start start with what is AI then we will see what are the different types of AI then we will see in this AI how that agent forces coming into picture after that we will see why we do need agents okay obviously we can do so much on automation then why we need agents after that we will see how the evolution of AI came into business then the five attribute utes of an agent and last will be the building blocks of agent force. Okay. So this is the agenda that we are going to cover in today's session. Now before I talk about the agenda, I ask you a very simple question that is what is AI? What is artificial intelligence? Any idea any anything that comes to your mind when you hear the term AI?

[00:05:58]Yes. No.

[00:06:06]Okay, you know what? When I heard this term AI for the very first time, this is what came into my mind.

[00:06:14]Now, if you see this diagram or if you see this drawing, what can you think when you see this drawing?

[00:06:23]Simple question. What are you able to see in this drawing or in this diagram?

[00:06:30]Let me call out their name. Kungran Mah Babu Tri V Karuna. Any idea what you're seeing?

[00:06:39]What what is are you able to understand when you see this AI or this diagram?

[00:06:48]Yes. Good. Please go ahead.

[00:06:50]>> Uh brain >> brain. Okay.

[00:06:54]>> Uh neurons brain neurons and all.

[00:06:58]>> Yeah. Okay. Yes. Yes. Correct. Anyone else?

[00:07:06]Evolution in all sectors. Okay. Great.

[00:07:09]Mahbu.

[00:07:10]Cool. So, you know what? When I was learning this AI for the very first time, I feel like whenever I learn anything new, I feel like connecting those things with anything that I'm seeing nearby to me or anything that I'm seeing in real uh real world. Okay. So when I was learning AI, I was finding it hard to understand actually what is AI.

[00:07:30]Okay, when I say when people when I open Google or when I ask LLMs, you know, JPD, it used to say it is an artificial intelligence and then a long definition.

[00:07:39]But it was very hard for me to connect.

[00:07:42]Okay. So what I did is I connected with my brain. So what does a brain does? So when you were born, what was the first thing that you did? When you were born, you made some moments and then you your parents ask you to you know say mama or say papa and then you started listening to them and then you made this those sound then you started walk crawling slowly you saw the surroundings okay you adapted yourself to that surrounding then you started walking then you started picking up things and now you are at a stage where you are running walking going to office working on the laptop of everything that you are doing is based on two things. What first thing is what are the external factors that making you do that. Second thing is what are the different types of data that you are trained on?

[00:08:34]Correct me if I'm wrong. Right? The data and the surroundings that you're seeing around is based on what you have evolved yourself. Right? In the same way it is called as so this particular thing is called as human intelligence. Now any machine or any system that works very much similar to what humans can do.

[00:08:56]Okay, which is not natural is called as artificial intelligence.

[00:09:01]Right? Now what does artific artificial intelligence do? Now if you see AI is everywhere this is all the different sectors that you are seeing. Okay.

[00:09:11]Real estate, e-commerce, finance, anywhere any sector you open up there will be an ear. Now what this AI is doing? This AI or artificial intelligence is working very much similar to what humans can do. Okay. And how it is working based on the different factors on which it is being trained.

[00:09:31]Number one and number two the different type of data that is being feeded into it. Correct? So isn't it very much similar to what humans are doing? Right?

[00:09:41]So if human is doing based on these factors they are called as human intelligence. But if machine is doing or if any system is doing then it is called as artificial intelligence.

[00:09:52]Pretty straightforward, pretty easy to understand. With thumbs up if you got it.

[00:09:58]Okay. So I will not take any bookish knowledge. I will take a realtime example and I'll explain it. Okay. So that is what artificial intelligence means according to me. Okay. Thank you GU and Kuman. Appreciate your reactions.

[00:10:13]Now we understood what is AI that okay now let's understand what are the different types of AI okay now if you open any LM any Google I mean any browser you will get three different types of AI one is called as predictive AI second is called as generative AI and the third one is called as agentic AI okay now before I talk about these different types of AI first you should understand the levels and stages of AI where you which you will not find even on your trail heading. Okay. Now there are basically three levels of okay what are those?

[00:11:00]I know I'm very good at drawing and you can see my drawing skills.

[00:11:07]So this is level one.

[00:11:10]So this is level one. This is level two and this is level three. Okay. Now in this level one is artificial natural intelligence. Level two is artificial general intelligence and level three is artificial super intelligence. Okay. Now you will be like what is this? These are the different levels of AI model or artificial intelligence. Okay. Now artificial natural intelligence means what? Artificial natural intelligence is a part of AI which will work lesser to what humans are doing.

[00:12:01]Okay. Like it will be able to perform the task which is lesser than what humans can do. Artificial intelligence, artificial general intelligence will be able to perform the task 90% close to what humans are doing in artificial general intelligence and artificial super intelligence will exceed the part which humans are doing. Now you will be like what does it mean? Right?

[00:12:28]So I will go little bit ahead and I will show you what I mean. So this three levels has basically five stages.

[00:12:40]Okay.

[00:12:44]Stage one will be your LLM.

[00:12:47]Stage two will be your agents.

[00:12:51]Stage three will be your multi- aents.

[00:12:57]Stage four will be your artificial super intelligence.

[00:13:03]Okay. And stage five will be TV which I don't want to talk with now.

[00:13:10]We will see it in the advanced part. So LLM means what? Stage one. What is LM?

[00:13:15]LMS are nothing but your Gemini claude your favorite GPD. What they does? What all they do? You put some inputs. Okay.

[00:13:29]What those inputs are called? I will not say that. you give give something in the text format and then it will generate the response that's what it does Gemini cla and chip right then it gets evolved and then there are some agents what these agents do it will do actions on your behalf for example what you will say hey draft an email so what it will do it will understand the context okay it will understand are the different things that has been performed and then it will send an email on your behalf. You don't need to worry about that. Multi- aents. Now what does multi- aents will do? There will be multiple agents. Agent one will be responsible for doing research. Agent two will be responsible for drafting the financial things. Okay. Let's suppose you are working on your finance of your company. Okay. Finance of your company.

[00:14:28]So agent one will do the R&D research.

[00:14:31]Agent two will draw the financial report and agent three will do the necessary stuff like it will go ahead and it will send an email. Okay, it will go ahead and clear the balance sheets. Okay, this all. So multiple agents are coordinating with each other and doing the parts which humans were doing. Then will be artificial super intelligence okay which is still in talk which is still not you know now it's still not live still not available but this is what is expected to go beyond what these all things are doing and it will exceed the human expectations okay now you'll be like then ammon what it has to do with this particular diagram that you have shown here so when I talk about this five stages out of this five stages three stages are where we have reached still here.

[00:15:29]Okay, we have already crossed artificial natural intelligence and we are in artificial general intelligence. Okay.

[00:15:36]Now you'll be like Ammon we understood okay there are three levels and five stages out of which you have shown us four stages. Then what about this this particular diagram that that you're showing as well. So in this three levels okay here is where you will find this three different types of AI. Now you'll be like Ammon then where was this predictive AI? Predictive AI was in artificial natural intelligence. So if I tell you what is predictive I will take a very basic example how many of you you have been using Netflix or let's say not Netflix let's say how many of you are being using Amazon Prime anyone here who use Netflix or Amazon Prime with thumbs up if you use it okay I see one person giving a thumbs up others you you might be using any other Right now tell me one thing based on your movies that you're watching or based on the shows that you're watching have you ever observed Netflix will show you movies or shows okay based on the type of movies that you or based on the different types of things that you're seeing on Netflix right based on your history that's what it will show that means what it is doing prediction based on what you are saying. Okay. Now that that is what a predictive model does. Okay. So this is predictive AI.

[00:17:12]Now if I say um if you will be like Ammon this is a good example but can you take any example which is in terms of Salesforce because we are now learning agent force which is again a part of Salesforce right now in this predictive area. How many of you have seen opportunity scoring or there is a concept called lead scoring? Einstein opportunity scoring, Einstein read scoring. Quick thumbs up. How many of you have heard about this term?

[00:17:40]Okay, again I can do others you haven't heard about it. See if you give me a reaction I it will be easy for me to understand. Okay, you are able to connect with what I'm talking about.

[00:17:57]Enabled it in or okay that's very good.

[00:18:00]Okay, Kaman says no, you have not seen it. Okay, so people if you have heard about it well and good. If you have not heard about it, I will take an example and I'll tell you. So Einstein opportunity scoring sorry Einstein opportunity scoring okay is what? So basically based on the different type of opportunities that you have worked on when I say you means your account executives have worked on based on that it will generate a score.

[00:18:33]Okay which will be from 1 to 90 you sometimes it might reach 200 as well.

[00:18:39]Okay now what does it mean? So whenever you will try to use this particular feature you know it will say hey you know what I need at least 400 opportunities okay in last 2 years.

[00:18:55]That means I need a data of 400 opportunities which have been utilized in last 2 years. Out of this 400 I want to see have data of 200 opportunities which was in closed loss stage and 200 opportunities which were in closed bond stage. Now your AI I will not talk about LM I'll say your AI okay your predictive model will get trained on this data okay now what will happen is whenever you're working once it gets trained on this data whenever you have new opportunity let's say the name of the opportunity is OP1 whenever you have this new opportunity okay it will analyze all those different factors like it will first see all the different fields what all the activities that you have performed like you have sent an email or you have scheduled a meeting. Okay. And other things all the factors that are there on that opportunity record detail page all those factors it will analyze and then it will generate a score. Let's say the score is 75. It will generate the score and then it will tell you the factors that is needed to win this opportunity. Now how this score got gener generated? It's a prediction. It's not accurate one. It is doing a prediction. So this prediction is done using this predictive model.

[00:20:25]Okay, that is called as predictive AI.

[00:20:29]So that means Salesforce already had this predictive AI. Okay, it was Einstein opportunity scoring or Einstein lead scoring. Same thing is there for Einstein lead scoring. Now after this there is a new concept that came up in the market from March 2023 September 2023 that is generative AI which you all like very well which is called as chad GP2. Okay GP stands for chad regenerated transformers. Okay.

[00:21:00]Now in this generative AI what it does it will generate text it will generate audio it will generate images it will generate videos right it will generate images based on the text input that you are given right so it will take the input now that input is called as prompt and why it is called as prompt that we will see in the next session that is tomorrow's session okay it will take that input and it is going to generate the response in the form of text audio video and images Right? So that is your generative AI. But these two are not enough. Why? Humans being humans they were like okay it can does a prediction.

[00:21:40]It can generate responses in the form of text, images, videos everything then why it cannot take an action for us right we want to just humans are like you know what we want to have a button.

[00:21:54]I will click that button and everything should happen. That's the thought process of humans.

[00:21:59]Okay. And that gave birth to your agentic AI. Now in this agentic AI, what will happen? The agents, there will be some agents who will take an action on your behalf. Who will do your the work on your behalf. Okay? Since I see most of the crowd is from India. I hope you all know in India there is something called as Tatkal booking, right? Tatkal ticket booking for Indian railways, right? So now for that we are so much busy that we do not get time to book our T thatal tickets right. So what we do we say that hey I will there is an agent who will take 500 rupees extra okay from the base fair of the ticket but he will or she will do the work for us like they will go ahead and book that particular train ticket. Okay. So that is what an agent does. In the same way these agents were born from AI as well. So those agents are called as AI agents. Okay.

[00:23:00]All clear. Any doubt to here? Quick thumbs up if you have everything clear.

[00:23:11]Okay. Thank you. I see.

[00:23:15]Yes. Lot of thumbs up from you. Great.

[00:23:18]So now since you understood this different types of AI. Okay, I purposely didn't show you this diagram which you will see on your trailer as well because this is very hard to understand. Okay, I took a very basic example and I showed you. So if I take off this diagram which you will see on the trailer as well.

[00:23:33]First it started with predictive, then it was co-pilots. Now what is co-pilot?

[00:23:37]It's nothing but generative air. Then now we are in the third wave which is agentic. Fourth wave will be robotics and after that it will be artificial general intelligence.

[00:23:47]Okay, same thing I explained you using this diagram. Okay, now next. Now you understood what is AI. You understood what are the different levels of AI. You understood what are the different stages of AI. You understood what are the different types of AI. Now it's time to understand agent force. What is agent force? Agent force is nothing but it is an agent. If I talk in terms of Salesforce, it is agent force is Salesforce enterprise platform that is being used to build customize and deploy agentic AI on this different business enterprise applications. That is what agent force is.

[00:24:37]Okay, make sense? Now if you open any agent force on the Salesforce website you will find this tagline humans with agents drive customer success. Why this particular tagline is coming? The reason is very simple. Agent force agents will be built by humans. Okay. Then it will be deployed on a third party system and there also human intervention is needed.

[00:25:01]Okay. So why we need agents? Basically we already had so if you remember in the beginning Mahesh Babu wrote that it's what when I asked him what is agent force agent he said it's a salesforce user support chatbot time. Now if I ask you why we need this agent if we already had this chat box so it is basically this agents would act as a bridge between your LLMs and the business use case. Now what is LLMs? So LLM stands for large language models. Okay, Gemini Claude or Chat GPT okay cursor anything that you open up those are nothing but LLMs. Now this LLMs work in a generic cause. Now your chat GP will not know okay you're working on Salesforce fine. So I need to be curated in a Salesforce way. No that particular LLM is being there for multiple type of task. It is not there only for a specific task. Right? So now this agents that you are going to bring build this agent force agent that you're going to build okay it will act as a bridge between your LLMs and the business use case. Okay for example I'll say that hey you know what I need an agent who can go ahead and let's take a very basic example of what Salesforce is doing every day. Coral cloud. So Coral Cloud Resort will have an agent which will go ahead and book the hotels. Then they will check whether there are any spa sessions available or not and it will compare the fly uh flight tickets, hotel tickets and then it will give you the best possible solution. Okay. So that is what a corl cloud agent is doing. Okay. Similarly, you will build an agent which will act as a bridge between your LMS and business use case.

[00:26:49]Now it is well organized under functional areas which is called as topics. Now topics has been renamed as a sub aents. Okay, which we'll see after some time. It orchestrates actions using LLM reasoning. What does it mean? So you know what inside the agent force there is a concept called atlas reasoning engine which is also called as the brain of agent force agents. Okay. So all the actions that your agents are going to perform okay it will happen based on the atlas reasoning engine. So that's what I have mentioned. It orchestrate action using the LM reasoning. It is grounded in enterprise data. Now what is grounding? Data grounding means. So that we will see later. But it is grounded in enterprise data means it has been fine- tuned in such a way that it will work only on the enterprise data that you are going to provide. It is not going to work on the generic data which is available on the website or any other place. Okay. Next, it is trained using guardrails. Now, what is guardrails?

[00:27:55]Guardrails means some rules and restrictions or the boundaries that has been set. So, whenever you build your agents, okay, which whenever you build your agent force agents, it will be working under the guards that you are setting up. For example, let's suppose Karuna is there. Okay, Karuna is Salesforce admin. So based on her profile the agent can go ahead and explore any data that agent wants right but let's take an example of goodu is a sales representative or he is an account executive now he is entitled to work only on opportunity accounts and contacts that's it he cannot go ahead and work on or cannot go ahead and see cases he cannot go ahead and work on the leads that has been generated or he cannot go ahead and send an email to those leads or XY Z any other task he cannot perform right now if you're building an agent on behalf of Budro okay it should follow those guards it should follow those restrictions rules that has been set so for that we have this guardils that agents will be trained on this guard and the most important thing if someone ask you you know what I have claw which will help me to build the agents. I have cursor which will which can work like an agent force v. Then why we need this agent force agent? The most important underlying statement is it is secured by Einstein trust level. If you build any particular agent there is no Einstein trust layer.

[00:29:32]Okay. What is Einstein trust layer and how it works? We will see that later.

[00:29:35]But that what that is what makes the agent force agents more secure more reliable as compared to the other agents that are available in the market.

[00:29:43]Okay. Now the most important question we already many people feel that this agents are or agent force agents are very much similar to chat bots. Now how many of you have built chat bots or how many of you have interacted with chat bots? Quick thumbs up if you have built or if you have interacted with chat bots. Okay. Gundu says gives a thumbs up others.

[00:30:10]Okay. None of you have built chatbot or none of you have interacted with chat bots.

[00:30:21]One, two. Okay. Tamran says he has interacted. Good. So whenever you interact with chatbot what it does, it is very repetitive. Like if you ask anything out of what it has been not been trained on. Thanks KPI. So if you ask something which which it has not been trained on, it will not give you a response. Right? Like I remember I used to use a chatbot which was again part of Deepa I don't know how many of you okay which is again part of your um Indr status it will show you okay if you ask what is the train schedule it will show you but if you ask like hey can you help me predict what are the chances of getting my ticket confirmed it will directly go ahead and create a case for you. Okay. Or it will say this is not in my context. So that means those particular chatbots that you are seeing can I say that they are repetitive in nature. Right? If you ask to do something different to it, it will not be able to do it. So it is repetitive in nature. Okay. It is scripted. If you write anything out of the script, if you ask anything out of the script, it will not be able to do it. And it provides very limited information. Right? Whereas when you go with this autonomous agents or agent force agent, it is dynamic in nature. Okay, if you ask something which is not there or which is not possible for it to do it. What it will go and do?

[00:31:50]It will go and try to read or understand the entire context. Let's suppose I'm talking about the confirmation chances of my ticket. So what it will do? It will go ahead and see the history of that particular train of confirmation chances, the speed or the demand of those particular tickets and then it will try to give a approximate answer but not the exact answer. Right? It is not repetitive but it is proactive in nature. It is conversational like you know why people are more towards agent force agents. Okay.

[00:32:30]Is because it is more conversational. It gives more personalized talk whenever you're talking to it. Okay. And the last one is information plus action label. So for example, I have built an agent or we this I have built it one and a half year back like if any of things which you are asking the agent is not able to answer.

[00:32:50]Okay. What it used to do is it used to understand all the entire context. etc. It was trying to help with the best possible answer but if it is not able to answer it used to directly route the using omni channel routing it used to route that particular request to a human agent and then those particular issue were taken care itself which was not possible inside the chat box okay that's how the evolution of AI came into business okay making sense everyone any question any doubt till Yes, we that's part of it. Thanks.

[00:33:35]Any other questions before I proceed ahead? I will continue for the next five minutes and then I'll open the floor for any doubts or any question that you have.

[00:33:45]No. All good.

[00:33:47]Can I get a thumbs up so that I understand? Okay. The crowd is with me or the audience is with me.

[00:33:55]Okay, thank you.

[00:33:58]Now we understood the usage of having this AI agents or autonomous agents, right? But the most important thing is the five attributes of a zoom. You whenever you are building any agent force agent or any type of agent whenever you are building it, you need to know what are those five attributes.

[00:34:18]First thing the role of an agent. Okay.

[00:34:22]Now, for example, I told you I took an example for Tatkal ticket agent. Okay.

[00:34:28]What is the role of that agent? That particular agent will book only the train tickets for Indian railways. It will not go ahead and book tickets for Pakistan railways. Right?

[00:34:42]It will not go ahead and book tickets for Russian railways. It will they it is designed that agent is role is to book tickets for train and that to for Indian railways that is the role of the agent right so in the same way whenever you are building any agent force agent you first need to tell okay okay we will see the practical example of this as well okay the role of that particular agent what that agent is designed to do okay now you'll be be like if I want to build an agent who can book tickets who can tell me what What is the status of my tickets that I have booked? What will be the prices that like you know the flight prices that is going to increase in future? So all these things I want my agent to do. Okay. So what will be the role? So again that will be your booking assistant agent.

[00:35:36]Okay. Now since you have defined the role, you need to tell the data on which this particular agent is going to deploy.

[00:35:44]Okay, for example, this particular agent is going to work on tickets. Okay, so it should get access only to the tickets data. It should get access only to the train data. Okay, that means which particular train is going to come. It should get access only to this particular data. It should not get access to account object, contact. Okay, maybe contact should be there because you are going to book a ticket for a particular customer but not related to an account or employee or any other XY Z. So that you have to take care that your agent is being designed or being configured in such a way that it is accessing only the limited data that it has to use. Once you know the role, once you tell the data, the third thing is actions. Which actions it needs to perform? whether it should send an email after the ticket has been confirmed. whether you should book the ticket okay whether it should show the pricing comparison based on the different websites that are available all those thing okay so that is what the action is now the next one is guard which I already told you what agent shouldn't do okay that means do not book the tickets do not go ahead and make a advanced payment do not cancel the ticket if I'm not traveling.

[00:37:07]So this kind of guidance you have to set. This kind of restrictions you have to set or rules you have to set that is guard rules.

[00:37:16]Okay. Next is channel. It's very important where your agents is going to sit. Whether it is going to be sitting on the website.

[00:37:26]Okay. Whether it is going to sit on the slack channel. Whether it is going to sit inside my Microsoft teams. whether it is going to sit inside my Salesforce or this you need to decide. Okay. So whenever you build any agent first and foremost thing is to identify the five attributes of an agent.

[00:37:46]Okay. Any any questions? Any doubt till here?

[00:37:50]Quick thumbs up if it was easy to understand.

[00:37:57]Yes. Okay. I see one thumbs up. Two.

[00:38:00]Three. Okay. Good. Now since you understood the five attributes of an agent, we will see what's next. So the next is the building blocks. There should be five attributes and three building blocks. Building block number one is topic which has been renamed as sub aents. Why it has been renamed that we will see later. So topic means the scenarios on which your particular agent is going to work on.

[00:38:25]Okay. So that is what your topic is or the sub agent fields like for example order status inquiry password reset these are a different topics based on this topic what will happen is your agents will go ahead and work on the instructions so yeah I'll come to your question so how will you remember this so you will remember this like this so this is your umbrella Okay. So this is your umbrella which is nothing but your sub aent. Sub aent means topic. Within this topic within this umbrella will be your instructions and actions. Okay. So this particular umbrella is topic. Within this topic there will be certain instructions. What your agent will do? You know how now you will be like how my agent is going to work. So your agent will first see the topic. So your atlas reasoning engine whenever you are giving inputs okay which is called as pron your atlas reasoning engine will go ahead and find out okay which particular topic I need to use. Once the topic is selected then it will select okay I need to go ahead and follow all these instructions. It will read all the instructions. Once it reads all the instructions the next thing that it will do is it will go ahead and execute that action. It will find out okay based on this instruction which will be the best action and then that action will be performed. Okay. So yes, do you have a doubt Ka?

[00:39:57]>> Uh yeah, you are talking about the agents, right? So uh like how we will be uh you know build this agents like other if we talk about others environment not in a sales source like we must have some prerequisite like we must know Python we must know every other's languages but how agents will work in the Salesforce is there any connectors only or if other languages we will be uh covering in the sales how we will be you know uh interact with the sales source because in a salesforce it is completely based on apex right so how they will be built like it's a only configuration or any technical expertise we we must need for building all the agents and all >> first of all uh for building an agent inside the agent for uh inside the agent force or let's say inside sales force it's not it's not tough Okay, it's all configuration.

[00:41:03]Okay, >> to use that configuration, you need to be good with your flows, AEX classes.

[00:41:10]Okay, these two things are important.

[00:41:12]Not to build the agent, but to use the actions. So there are already some standard actions and then there will be some action that you will be building.

[00:41:19]Okay, that to use that actions, you should be good with AEX and close. Okay, >> there is something new which we will see at the last of this training which is called agent for script. Now if you're someone who don't like to use configuration you're someone who was like yeah man I want my agent to be very determinist okay in that situation you will use agent for script that all we will see but yes just an overview I'm telling you these are the different ways where how you can build the agents >> okay so it's a mixup like some other way some configuration and some other ways in a declarative form okay >> yes >> okay fine thank you >> has given both the options. It's on you.

[00:42:02]You want to go ahead with which one?

[00:42:07]Okay. Any other questions?

[00:42:10]>> It's a completely based on business logic like what business required, right?

[00:42:15]>> No, no, no. I'm telling you the different ways of building an agent.

[00:42:19]Salesforce has given you.

[00:42:22]>> You can do it using drag and drop. You can you do it using by writing scripts as well. Both the ways are available if you want to build a reason.

[00:42:30]>> Okay. Okay. Fine. Okay. Thank you.

[00:42:33]>> Yeah.

[00:42:35]Any other questions anyone?

[00:42:40]Okay. I'll take it as no.

[00:42:43]So what are the key components and how this agents work that we need to see?

[00:42:48]Now I was telling right this is very important step.

[00:42:51]First the triggering point. So the triggering point can be through a conversation or through the data that you are accessing. Okay. Or through any automation that is going to work. So this is the this are the different triggering point for an agent to get started. Okay. That means let's take an basic example that you have given some input which is called a spr. Now that context will be sent to the LLMs. Now you'll be like that means when you talk about LLMs my data is going to go outside Salesforce. Yes. Salesforce says yes your data will be be going outside Salesforce. Now you'll be like that is a very big security bridge how my customers will be very much confident that their data is very much secure. So for that your Einstein trust layer comes into which we will see in the next session.

[00:43:43]Okay. But here I will show you. So your context is sent to the LLM and if the LLM is not able to understand very much clear with what you have given. So it will ask you to clarify the intent again and again it will ask you the questions.

[00:43:59]Once it understands everything your atlas reasoning engine the brain will come into picture which will go ahead and select the topic and instructions.

[00:44:07]Once the topic and instructions has been selected the chaining of actions will get like all the actions will get executed properly. Now let's suppose there was an action to query records.

[00:44:19]That action got executed. Again it will go ahead and pass this response. Okay.

[00:44:25]Now it will understand okay now I have got the records. Next thing I have to go ahead is I have to send an email to this records which I have received. But as per the instructions am I allowed to do that? As per the instructions this particular agent does have access to that. If it is there then it will go ahead and execute that particular action. So chaining of actions and this selection of topic and instructions will keep happening in a repetitive manner.

[00:44:54]After both of this has happened then it will get gather all that input and it will execute that action. Once that action has been performed it will give a confirmation to the user.

[00:45:07]Now I know very well that this was very layman terms I've explained you but you need to wait for tomorrow's session to see this particular things in action okay right from the scratch so with this we will move to the last part okay here you have to create a free developer or I know you might already you already have a new or if not please create a new org with this particular link.

[00:45:41]Okay, I'm pasting this link that's a very simple link like salesforce developer.salforce.com/signup.

[00:45:49]I'm pasting this link in the chat. Okay, please create a new developer or even if you have a developer or I would request you to create a all together a new org.

[00:46:02]Okay. And second thing is please complete this trail head so that tomorrow once you come you understand with what I have explained you today.

[00:46:12]I'm again placing this link also in the chat.

[00:46:16]So this task I'm giving you for today and I will leave you with this particular video. This is a very beautiful video that I want you to see before you go. So let me share my screen.

[00:46:40]Let me know when you're able to see my screen.

[00:46:47]So are you able to see my screen?

[00:46:53]Yes. Yes. Yes.

[00:46:59]Let me and also the audio if you're able to hear it let me know. Okay.

[00:47:14]>> It took me years to learn.

[00:47:16]>> Are you able to hear the audio?

[00:47:20]>> Yes.

[00:47:23]>> I'll go on mute and I will Ask me to enjoy this audio video.

[00:47:29]>> Years. It took me years to learn this.

[00:47:31]And now now anyone can just type a sentence.

[00:47:38]>> I'm sorry, Chloe. It's just more efficient this way. The robot makes a perfect cup every time. But I love working here.

[00:47:45]>> They don't need me anymore. AI can create stunning visuals now. Cheaper, too. It's over.

[00:47:53]Well, he always remembers our anniversary, and he never complains about my cooking. After three human husbands, that's a refreshing change.

[00:48:01]>> Now, my camera just gathers dust. A single prompt and the AI creates what took me a lifetime to learn.

[00:48:09]>> I used to design logos, websites, you know. Now, some program does it in seconds. Funny how the future works, isn't it? Here.

[00:48:19]What is the So, what's your take on that video?

[00:48:31]Found it funny. Interesting.

[00:48:37]Anyone?

[00:48:45]Did you all saw that video or see that video or No.

[00:48:53]Okay. Anyways, the next thing that I wanted to tell you is that video does look scary but AI will not replace humans but but but it will definitely outsmart people. So AI skill humans will definitely outsmart others. With that said, thank you.

[00:49:18]So now open open the floor for your any kind of questions that comes to your mind.

[00:49:30]I see a lot of tension coming up.

[00:49:34]Any questions? Anything before we wind up for the day?

[00:49:46]All good.

[00:49:48]Okay. Cool. Then thank you so much everyone. Baby says I think we are good to close then.

[00:49:54]>> Thank you.

[00:49:55]>> Yeah. Thank you and to all I will have the recording ready in few minute to everyone. Okay.

[00:50:06]>> Thank you so much.