What was the need for this?
What was the need for this? Let’s understand. So, first, let me explain it to you in the picture. Yes. It’s like a box. Okay?
The world of AI
So understand that this is the world of AI. It’s the world of AI. Now, in this, the issue is that LLM was brought in first.
Large Language Model
Now you might ask, “What is LLM, Soumya Di?” So, the full form of LLM is the Large Language Model. This is a Large Language Model.
ChatGPT, Gemini, all these things
If I give you a very simple example, I’ll take my ChatGPT, Gemini, all these things we use. Like, we asked it to write a paragraph for us.
Write a LinkedIn post
A LinkedIn post. Write it down, and it’ll write it down and give it to us. Right? If you haven’t used ChatGPT and all that, then that’s no problem.
I’ll demonstrate right now
I’ll demonstrate right now what it is and how it works. Okay? So, basically, LLM, regardless of GPT, Geminis, call these people LLMs.
Large Language Models
Large Language Models are available for whatever you need them to do. Now, all these people can’t do the work. That’s their problem.
Now, all these people can’t do the work
We’ll understand. Now let me explain to you what they can’t do and why. You can. Now, if you give an LLM some simple instructions, like, “Brother, write a LinkedIn post for me, or a blog post for me.”
Write a quick email
Write a quick email. Right? Just do this assignment for me. He’ll happily do it, smiling and laughing. There’s no problem with that.
They can understand natural language
You can see it says, “They can understand natural language, answer your question,” right? Code, explain the concept.
A very intelligent text generator
But the most important problem was that LLM is a very intelligent text generator, but it’s not a great worker. If you want, you can automate things with it.
Schedule a meeting
Okay? Paul Robinson was my manager at my last company. He needs to email us to say he wants to have a meeting.
Please tell us the time. He’s already told us the time. Now, if I want to know what time did Paul Robinson say? Schedule an invitation on your calendar for that.
Schedule a meeting. Or maybe it’s not that difficult. What if I ask him what time my meeting is? Like, read my email and tell me what time the meeting is?
What the weather will be like during the meeting
Or what the weather will be like during the meeting? That’s nothing. It won’t be able to do anything. Why won’t it? Because it doesn’t have access.
It doesn’t have access
What doesn’t it have? It doesn’t have access. The biggest problem arose with LLM, whether with ChatGPT or then you can choose any other model.
The ChatGPT has become the most famous
ChatGPT has become the most famous. So that’s why I’m talking about ChatGPT. Okay? Now if you ask me, “Can you tell me what the meeting is by looking at the calendar? What time is it?”
The biggest issue
Now, here’s the biggest issue. The problem was access. Okay? Now let me explain it to you a little more.
Limitations of LLM
Limitations of LLM. Once this theory is understood, we’ll do it. Then we’ll demonstrate it practically so there’s no doubt left in your mind. Okay?
An LLM cannot take action
The limitation of an LLM is that, brother, an LLM cannot take action. It can only tell you how to do something, but in your place, he can do it.
I can’t. Okay? For example, it can explain how to book a flight. But it automatically loads the browser.
I can’t open it, compare prices, and book it for you. Right? That was the problem here. In a way, the LLM is a brain without hands.
It has a brain that can tell you things, but it can’t do them for you. It doesn’t have the ability to do real-world things. I don’t have access to it.
A brain without hands
Like the example I gave you, if I ask it, “When is my meeting with Paul Vinson?” It doesn’t have any access to the calendar vendor, so how can it tell? Where can I see it?
It can only generate text based on training data
Okay? So this is the best thing we have. There was a big problem: It can only generate text based on training data.
It has no access to your personal environment.
For example, let’s just take a quick walk and see
For example, let’s just take a quick walk and see. Okay? Why just talk? We’ll figure it out.
Please write an email
If I write, “Please write an email.” What should I do? Please write an email. Asking Paul for meeting time. Okay?
Write this email well
That’ll do. No problem. Write this email well in it. Now let me ask him, can you tell me when the meeting is? What did Paul reply? Isn’t it?
I want full automation
If I want full automation, it’s saying I don’t have a reply. And how can I get a reply? He doesn’t even have access.
When is my meeting?
If I ask him now , when you ask directly, “When is my meeting?” Often, the person will simply tell you, “When is my meeting?” I don’t have access, man. Okay?
There is no confirmed schedule
Right now they’re just saying, “Tell me your preferred time slot.” There is no confirmed schedule. Can you check in on the calendar? Okay?
I can’t directly add it to your calendar
If we say it like this, I can’t directly add it to your calendar because of access issues. Okay?
ChatGPT will basically help you with many things
So, ChatGPT will basically help you with many things. It will definitely make things easier for you. It will generate answers for you.
There are some things you can expect from it
But there are some things you can expect from it. You can have a calendar, email from another app, or you can just write and send it. Automation, basically.
This won’t be possible. Okay? So now we understand that the brother who was an LLM, the problem with GPT was that it was, in a way, a brain without hands.
They created something called an AI workflow
Now the human mind is fine. Now it thinks about what’s better. So what did humans do? They created something called an AI workflow.
We started granting access accordingly
What they did with this AI workflow was that the LLM program, sir, was LLM as per your need. We started granting access accordingly.
We’ll give you access
For example, if we wanted our LLM, automate everything from the calendar to the meeting. So, if I want to schedule a meeting and tell it when it’s scheduled, I’ll use the Calendar app.
We’ll give you access. Okay? If I want to get the weather, etc., we’ll give you access to the Weather API.
We started giving access
If I want it to tell me things by speaking, then audio means API of the app, etc. We started giving access so that the LLM students who were brainless without hands could, we folded our hands to him and said, “Brother, you have hands. Now you are ours.”
Provide us with a great job done in a fun way. But even here, what problem do humans have that somewhere or the other our brains are being used? The human brain is being used.
What access should be granted
Because what does LLM sir need? What access should be granted, what access should be denied? Humans have to figure out all this.
Humans have to think about it
Now, humans have to think about it. We need to do this in a way that we don’t even have to think about what we should give access to and what we shouldn’t.
AI workflow
So, let me explain a little bit about the AI workflow. Okay? Otherwise, you’ll say the course is being rushed through too quickly.
What is an AI workflow?
So what is an AI workflow, essentially? This is it. Giving LLM access is an AI workflow. We understand that LLM generates content, etc.
LLM with Access is a workflow
But he doesn’t have access to the apps, so he can’t do things on our behalf. Right? So we humans decided we’d start giving it access to apps.
So, LLM with Access is a workflow. It’s clear. So much? The AI LLM is clear to you. Right?
We named it AI Workflow
And we’ve started giving access to the AI workflow, basically the LLM. We did it. We had access to whatever we needed to do, and we named it AI Workflow.
What a name! Dude? I’ve named it an AI workflow.
Developers started giving LLM tools access
Now, solve it. Developers started giving LLM tools access. Examples include the Calendar API, Weather API, Email.
I gave him access to the database, saying, “Brother, you can take all the access. Just do our work for us. Okay? So lazy.”
The system can read calendars, fetch weather, and send emails
Humans are becoming more and more. Are you seeing that? Now the system can read calendars, fetch weather, and send emails on your behalf.
Basically, the tasks that people are unable to do due to laziness. AI is now handing over all that workflow to humans.
AI workflows, completely automating their work
So, many people are still in the middle of their lives. You may have recently heard that some companies were given AI workflows, completely automating their work.
Millions and crores were charged in this way. There are many reports coming in. If you’re not receiving such news, your feed is messed up.
That’s not what people are following. Okay? And at this stage, we moved from pure LLM to AI workflow. Now we’ve moved from LLM to AI workflow.
The AI workflow is a structured pipeline
Where we’re giving LLM access to automate things. Right? And the AI workflow is a structured pipeline. Where the LLM is connected to tools.
The human decides which tool to connect. Right? Steps are predefined. For example, the user will ask some questions.
There is an LLM, then the LLM that has been given access will make the call. Calendar API or whatever, and then it will give us the output.
For example, if I’m asking when my meeting is with Paul, if he’s been given access to the calendar, so by checking it, it will tell you what time your meeting is on that day.
Why did we need AI agents?
Okay? Okay. But what were the limitations in AI workflows? Let’s understand that. Why did we need AI agents?
Their hierarchy is that first comes the LLM, then comes the AI workflow officer. There was a shortage. Access was an issue.
By giving these people access, we began building AI workflows. What was lacking in them that we needed to do? AI agents were needed.
Humans are still the Decision maker
I’ll tell you what was missing. So, even now, humans haven’t felt like, “Somewhere, humans are still the…” Decision maker.
We have to tell the LLM, “Here’s your calendar access. Here’s your email access, brother.” Access to the weather API.
So, human involvement was still quite high, right? [nasal noise] Workflow: Humans decide which tools will be connected and which won’t.
We’re the ones who put logic into it. Explain how you’ll be doing what you’re doing. Right? Within the workflow.
Humans define the order of steps. So, somewhere, humans, the environment was too much. And humans realized they needed something better, something that didn’t involve humans at all.
Everything should be completely automated. And so, Humans created the AI agent, The Suspicion.
The AI agent is at the top
And for now, the AI agent is at the top. What’s missing now? And it’s not known what humans will do to make it what it is.
But for now, an agent is the latest and one of the best AI inventions through which humans’ work is done in very simple ways.
Is that okay, sir? Are you still understanding things? If you are, then comment below with your LLM AI Workflow Sorted. Okay?
You’re not sitting right in front of me
Now, you guys. You’re not sitting right in front of me so I can see who’s nodding off, who’s eating a burger, or whatever. Okay?
So you, I get comments only through which I know that you people are able to understand. Are you unable to understand things? You’re ignoring it, it’s all superficial.
I hope you can understand. I’m explaining in very simple terms what LLM is? AI. What’s the workflow? My subscribers are incredibly intelligent.
And let me know which of you likes it. Have you already taken a Python course or not? Have you already taken my Python course or not? Okay?
Because I want to know. People who did that are doing this, or different people are doing this course. Okay?
Let’s get to the AI agent
Alright. So I hope I’m more. I’m not asking questions, friend. If I’m teaching you this much, then I have the right to ask you who my subscribers are.
Who’s studying? Right? What are they doing? Okay? Just talking won’t do anymore. Let’s get to the AI agent.
