AI Agents, Clearly Explained

Added: 2026-05-10

[00:00:00]So, right now there is a lot of hype about the AI agents. In this video, I want to talk about a simple way to understand AI agents and how you can easily understand what an AI agent is.

[00:00:10]How is it different from the normal chat that you do in ChatGPT? So, before we explain or before we go into the agents, first let's understand the difference between an AI app versus an AI model.

[00:00:22]And first you need to understand that ChatGPT and GPT-5 are not the same. So, ChatGPT is the app and GPT-5 is the model. So, they are not the same thing.

[00:00:31]The app is what you see in the front end like this which has a text box where you can type and you can see the previous conversation. So, this is the app.

[00:00:39]Whereas, the model is what actually thinks and what actually responds to you. And here you can see this is you and the app. The app is like a doorway or a gateway to the AI model or the LLM.

[00:00:51]So, you're basically going into the door and you're talking to the AI model.

[00:00:56]And these are the different apps that are there and these are the different AI models. So, for example, the ChatGPT is the app and then the GPT-4 or the GPT-5 is actually the model. And Gemini is the app and Gemini 3 or Gemini 1.5 and the Claude is the app and then the Sonnet 4.5 is the model. All right. So, next you can also think of it as the app as a car and the model is the engine that's actually the driving the wheels.

[00:01:22]So, this is a simple way to think about what is an AI model. And next, whatever question you ask it goes into the AI model and then you get the answer. So, the AI model is the one that actually generating the response. Now that we have understood the difference between the AI app and the AI model, let's understand how the AI agent works. So, this how a regular conversation works.

[00:01:44]So, you're the person who is going to chat with the LLM. So, whenever you give an input, the LLM is going to respond with an output. Basically, you give a text input and you receive a text output. So, this is how it works. It's very simple. Now, the main limitation here is if you don't give anything, the LLM is sleeping. It's not going to do anything.

[00:02:04]So, only time it responds is when you give an input, it responds.

[00:02:08]So, the first limitation is that you always need to give an input for the LLM to respond. Now, the next limitation is that there is a knowledge gap. So, the LLM was trained on the internet at a certain date. The model was trained few years back, and it only knows whatever information was there in the internet at that time. So, it doesn't have the latest information. Let's say, if you ask about what happened yesterday, it will not know. And it doesn't know about your email or your calendar or your files that you have on your computer.

[00:02:40]So, because it was trained on the internet, it doesn't know your world.

[00:02:44]So, these are the two fundamental limitations where it doesn't know your things or the latest information, and then you always needs to prompt it in order to get an output. So, let's see how a AI agent is solving this. Let's say you go to an LLM or AI model and you ask what's on my calendar. It is going to say, "Sorry, I don't have access to that. I don't know what is on your calendar."

[00:03:05]But, now, in order to make this LLM into an AI agent, you simply need to give it access to a tool. And now, we ask the same question.

[00:03:14]What's on my calendar? Now, the LLM is going to go back into the calendar and it's going to ask, "Hey, calendar tool, can you get the events for the week?"

[00:03:24]So, now, the LLM is calling the tool and then asking for the events for the week.

[00:03:31]Next, let's see what happens, okay?

[00:03:33]Now, here you go. These are the events for the next week. So, team standup, project review, client call, and this is the reply the calendar tool is giving to the LLM.

[00:03:42]And now, the LLM got the information from the calendar tool. It's going to reply back to you saying that these are the events for the next week.

[00:03:50]So, you can see how instead of you giving the input to the LLM every time, the LLM is actually calling the calendar tool, and then the calendar tool is actually giving the response to the LLM, and then it's responding back to you.

[00:04:03]So, this is the fundamental difference between a normal chatting versus an AI agent, where the LLM is having access to different tools, and it's able to get answers from that tool, and it's able to reply back to you. So, let's do a quick demo of this on a particular AI app like Claude. I'm going to use Claude, but it works similar across all the AI apps like ChatGPT or Gemini. So, let's see a quick demo of what we just saw. Similar to the message that we sent before, what is on my calendar next week? So, let's send So, I'm on the Claude app, and we are using the Sonnet 4.6 model. And right now, I have not connected any tool to this model.

[00:04:44]So, when I go into connector, you can see the Google Calendar is turned off, which means the model does not have access to this tool. So, now let's send it, and let's see what happens. So, like we saw, it's saying, "I do not have access to your calendar, so I cannot reply to this." So, now let's do another chat, and let's send the same message, but now let's turn on the calendar tool.

[00:05:07]So, I just turned it on, the calendar tool, which means now the model is acting like an agent. So, that's the difference between a regular chat and an agent, where agent is able to do things on behalf of us. Okay? It's acting as an agent for us.

[00:05:23]So, now when we see the connector is on, the Google Calendar is on, and now when I click on this, so, as you can see, it's able to answer it, but before it actually produce the result, let's see what it has done, okay? So, first it is making a request to the tool saying that, "I need to find the calendar events. The start time is this, the end time is this, and then order by the start time.

[00:05:50]So this is the response given by the Google Calendar where it says these are the events. So here you can see the event, the date and time, and you can see the name of the event. It's basically a list of all the events from my calendar. So once it received this response from Google Calendar, now it's actually producing the answer. So this is the fundamental difference between a regular AI chart versus an AI acting as an agent where it has access to multiple tools enabled.

[00:06:20]So now to add connectors or to add tools into your AI model, all you have to do is just click on this, and then inside connector just click on add connector.

[00:06:28]So now you will get different list of tools that you can connect to your AI model. So it's up to you how many tools you want to connect. And once you connect these tools, each tool will have its own set of capabilities. So let's do something like this, Google Drive. When we click on Google Drive, it has access to seven different tools, which means create file, download file, get file metadata, search files, list recent files, read file content. So all of these are different capabilities or different tools that the model will get access to once you connect this Google Drive. So these are all pre-built apps that you have access to. You can also connect your own apps that we'll see later in another video. So this was the demo. So let's go back to the slides and we'll look at another example. So now we got the reply back from Claude after it accessed my calendar. So now let's get back to the calendar and see whether these events are actually there.

[00:07:24]So I'm going to go back to my calendar, and you can see on the 11th of May we have client meeting with Sam.

[00:07:30]And that's there. And then we have on 13th we have the conduct webinar. So it got the same events from the calendar.

[00:07:39]Okay? So it's not just the calendar tool that you can give access to LLMs. You can give them access to any different tools. You can have multiple tools connected to an LLM. That's when it can do lot more stuff or lot more things for you.

[00:07:53]So here, let's see another example where we connect lot of different tools, Google Drive, Telegram, Gmail, Calendar, Slack, WhatsApp, Notion.

[00:08:02]What happens when you connect multiple tools to the LLM? So once the LLM has access to multiple tools, you can just ask it to do things like send an email to John. And the LLM can access the tool that is required to do that action. So for example, when you say send an email to John, it can access the Gmail tool, write a message to John saying whatever was your message. So you can say send a message to John regarding so and so. And then it's going to write the email, and then it's going to go ahead and then create that email draft in Gmail. Okay?

[00:08:36]And then it's going to come and tell you again that the email has been drafted.

[00:08:40]So you can see the email has been sent successfully to John. So what happened here is you give the action, the LLM talks with the tool and does the action, and then tells you that the action has been done. Okay?

[00:08:55]So this is the workflow. It's not just like one tool you can use. You can actually chain multiple different tools.

[00:09:01]So you can say send the mail to John and then schedule the event in my calendar.

[00:09:06]So you can give multi-step actions like this, and the LLM is going to first use the Gmail tool to send the email, and then it's going to use the calendar tool to schedule your event. So it can chain or take multi-actions as well. All right? So this is a huge advantage when you think how you can interconnect between these apps and then use a lot of things. So you can also say read my knowledge base or an article that I have in Notion, and based on that reply to John, and then schedule the event. So you can do multi-step events as well.

[00:09:38]So, which where it becomes more powerful. So, to recap, you can add your tools to your AI model, and you can give requests where the AI model does things along with the apps on your behalf. So, this is where it is acting as an agent.

[00:09:51]All right? So, I'm going to show you another demo of how the agent can use tools to get recent information. I'm going to show you that demo in a while.

[00:10:02]But, before that, let's understand how the chat really works. It's very important to understand how the AI is able to produce the output in order to really take advantage of the agentic capabilities. So, what I'm going to do now is we are going to see how the AI chat really works.

[00:10:20]Now, let's look at how the AI model is able to answer our questions and what are the things that it actually remembers about us and what it doesn't remember about us. So, let's say you send a message to the model saying that, "Hi." Okay?

[00:10:36]And it's going to reply back with a "Hi."

[00:10:40]Now, let's say in the conversation, I'm going to again say, "My name is Jack." And the AI is going to reply back, "Nice to meet you, Jack."

[00:10:50]Now, in the next message, I'm going to ask, "What is my name?" And the AI is going to reply back, "Your name is Jack." But, when this message, "What is my name?" is being sent, how does the model able to recall your name? That is because whenever you're doing this chat, the actual application is remembering your conversation history, and it is sending the entire conversation back to the model.

[00:11:17]So, when you ask, "What is my name?" not just this conversation alone is not going back to the model, but the entire conversation is going back to the model.

[00:11:26]So, that's very important to understand.

[00:11:28]And because the model sees that in the conversation history, you have mentioned your name as Jack. It is able to respond with the right answer. So, basically, this is called as the context. Let's see a live demo of this that we are able to understand it better.

[00:11:45]So, now we are on a website called as platform.cloud.com, which is actually a developer platform where we can directly talk to the model. We have full control over the conversation history, and we do not have the app automatically maintain the history, but we can actually control the conversation history that goes. So, I'm going to do the same demo over here where right now we are on the chatting with the Claude Sonnet 4.6 model.

[00:12:10]And here's where you put your prompt.

[00:12:12]So, I'm just going to say, "Hi." All right?

[00:12:15]And I'm going to run this. And when I run, only this message alone goes back to the AI, and I'm able to see the response.

[00:12:25]So, now what I will do is I'm going to add this conversation back into the message, and now I'm going to say, "My name is Shyam. What is your name?"

[00:12:38]Now, what I've done is initially, I just sent this hi message, and the model actually produced this response.

[00:12:45]I inserted the model's response into the chat conversation, and then I added another response, and now I'm going to run this. The model is producing a response. Now, I'm going to add it back to the conversation. So, this is the thing that the AI apps are doing by default, adding to conversation. So, now I have added it to the conversation, and now when I ask, "What is my name?"

[00:13:10]you can see the AI model is having access to all of this information right here, and based on this, it is going to answer my latest question or my latest message.

[00:13:21]So, now when I click run, it has access to the entire context, and then it's going to respond based off that. Okay, so this is the power of this. So now what I'm going to do is I'm going to remove whatever name I have mentioned, I'm going to remove that and then I'm going to ask what is my name? So now only this conversation history is going to be sent to the AI model and when I click run, you can see I don't know your name, you have not me told me at would you like to share it?

[00:13:50]So basically whatever is in the conversation history it is being remembered and it's being sent back to the model and then the model is just completing the last message. So now that we know this, let's look at another example where I ask the model about something which it does not know about, okay? So we saw before that the LLM was trained on the internet, but it has a cutoff date a certain few years back or a one year back when its training data has a cutoff. So it doesn't have any knowledge after that cutoff date. So to show this to you, I'm going to ask a question which happened recently, okay?

[00:14:29]So this happened 2 days back, our state election happened 2 days back. Let's see what the AI model responds when I ask that question. Who won the Tamil Nadu election in 2026? So when I run this prompt, you can see that it says my knowledge cutoff is early 2025 and I don't have the information about that.

[00:14:48]Here you can see the raw model does not have any recent information, it does not even have the tools to get the recent Now, let's add the tool so that it can do a web search, okay? Now similar to how we saw the calendar tool works where the AI model goes and asks the calendar tool to give its information, we can also add a web search tool to this LLM.

[00:15:13]So we are talking with the Claude Sonnet model and now I'm going to add a web search tool to this model.

[00:15:19]So to add the web search tool, I'm just going to click on tools and then I'm going to click on this web search. So, I'm not going to do any of these changes. I'm just going to click on add tool.

[00:15:31]And the tool is added. Let's close this.

[00:15:34]Now, I'm going to leave the same question as it is with the tool. So, as you can see, it now has access to one tool. So, I'm going to run the same prompt again with access to this tool. So, you can see previously it said I don't have access to this information. Now, let's see what happens when we run this. So, as you can see, it's first doing a web search and it was able to find the results in the web search. And then with this conversation history, it is able to produce the correct answer. So, because of adding a tool, it's able to run or use the tool, get the results.

[00:16:12]And those results are getting added into the context. So, that's very important to note. As we saw, like when we are adding to conversation, whatever is happening, all of the information from the web is getting added into the context and it is able to produce a response from that context.

[00:16:29]Okay? So, this is how an AI agent works.

[00:16:33]A regular AI model becomes an AI agent as soon as you give access to a tool.

[00:16:38]So, this is a simple explanation of how AI agent works and I hope this is very clear to you.

[00:16:45]And it made you understand the power of AI agents and how you can connect basically any different tool into an agent and the AI will call the tool whenever it is needed. So, now quickly I want to talk about a webinar that I'm doing every Tuesday and Saturday at 10:00 a.m. where I teach you the complete AI skills that you need, how to automate the tasks in your business or in your work. So, if you want to join this workshop, just click the link in the description and it'll take you to this page and then just register for this free workshop. All right? So, thanks a lot for watching this video. I will see you in the next video. Thank you. Bye-bye.