So this is a very, simple video where I'll try to explain, how the AI systems are changing or have already changed from basic chatbots to now AI agents.
00:11
2025 has been all about AI agents, and they are everywhere.
00:15
That what exactly means when we say AI agent?
00:19
What.
00:19
What's the technology behind it?
00:22
most of the AI chatbots, so from 2018 till, till 2024 last year, most AI products were just, a chatbots wearing different clothes.
00:32
So, for example, they could answer a question, they could summarize a text, they could following a simple prompt template and staying inside just one conversation tone.
00:42
That's the entire concept of a chatbot, or what we say, chat flow.
00:47
now what has changed is.
00:48
And what, you know what the limitations were there in that kind of system.
00:53
there was no long form contextual memory in it.
00:56
then there was no decision making, there was no looping inside that.
00:59
There was also, no kind of tool usage.
01:03
the LLMs were not efficient enough.
01:06
They did not have that kind of power to execute tasks on their behalf.
01:10
So, in short, we can see that, in simple chat flows, in those linear chat flows, they were basic, you know, these were an input was coming in and then it was static enough, and then an output was create, going forward.
01:22
and those kind of chatbots did not have that kind of dynamic agent flow, they did not have that kind of decision making they did not have that kind of adaptive, you know, learning capabilities or strategic capabilities that it could perform tasks on your behalf.
01:37
this, this was the core need for creating AI agents.
01:41
So I wanted to take a step forward and visually explain what's the difference between, chatbot and an agent.
01:47
and I got this creative, analogy which thanks to Google's VO3 technology, I was able to create one.
01:54
So on your left you are seeing a straight path.
01:56
It's a basic road, minimal information.
02:00
you're just going from one point to another.
02:02
It's just one flow of input.
02:04
That's how we can call it.
02:05
And this is how we would say it's linear flow, chatbot, chat flow, whatever this technology is.
02:11
Now, on the other hand, we are seeing a complex city structure which is loaded with information.
02:17
There is an entire map with tons of roads and branches and groups going around.
02:23
and on the center of it we are seeing like, you can consider this as a big AI navigation center.
02:29
That is the brain of it, that is making all the decisions and it is strategizing, all the, all the next steps, all the complex workflows.
02:36
now this tower, let's say it has all the access to the entire city structure and it can access all of it.
02:43
And then based on the use case, it can just access any information or any specific part.
02:48
And then using all the strategizing and decision making and tool usage information, it is then navigating through this complex structure.
02:55
This is, and, and it's learning on the go.
02:58
So this is how we can imagine this as an AI agent or an adaptive decision making agent flow.
03:04
So now coming towards the practicality side of it, how we can actually implement chat flows and agent flows within our organizations.
03:11
So luckily in answer agent we have both the options available.
03:14
We can create our basic chatbots, chat flows, whatever you want to say it.
03:18
And we can also create our agents that are dynamic and adaptive in learning.
03:22
so we have this entire documentation available.
03:24
We learn about all the details of what options we have under the chat flows, how we can create one, and then on the other hand we have also the option to create detailed agent flows that are able to process multiple looping, iteration, different complex routing structures, chaining, and all of that.
03:44
but I want to keep it very simple.
03:46
so let's say I am in answer agent and from the chat flows I've created this basic one.
03:53
So how I've created it, from the simple drag and drop I got an LLM node which is very easy to add.
04:01
1.
04:02
I got this LLM chain.
04:03
Then I connected a basic OpenAI chat model which is again since it is built on flow wise, it's very simple to connect.
04:11
then for the chat prompt template I gave it simple instruction like you have.
04:16
for the simplicity.
04:17
This is like a translator, chat flow.
04:20
Now again this is one simple linear flow of information.
04:23
you are giving text in one language, you are getting it converted it in the other.
04:26
So this would be considered as a basic chatbot, not an agent at the moment.
04:30
so let's say if we, if we test it right now and I ask it to, and, and I've already hard coded it that translate this specific, you know, input language into French.
04:40
So if I ask this like translate this one, this text, it would give me the translated version in French.
04:46
So that's how a simple chatbot is.
04:48
Now good examples of this would be any social media caption generator you are giving in transcript, it is then processing that transcript and then simply giving you a caption out of it.
04:58
Another good example would be an email writer, summarizer.
05:01
So these are all examples of basic chatbots, chat flows.
05:05
Let's see what agent flows can do.
05:07
All right, so to explain the difference between agent flows and chat flows, I think this would serve the purpose.
05:13
So to create any agent flow I can simply go to agent flow.
05:17
Under this one I can, you know, let me just walk you through so we can go in, we can add a new one or you know, you can through through your own prompt.
05:27
You can generate one by you know, describing what you want in your agent and you can select a model.
05:32
or else you can simply select all the different options available here.
05:36
whatever your use case is and, and whatever your plan is how to build an agent.
05:40
For my use case I was I wanted to create a caption writer with sort of like research capabilities.
05:46
So I can explain the difference that it is dynamically using the tools other than just you like processing a textual information.
05:53
So that is why I am using a start Note that I have provided it a tool.
05:58
and for doing the research I have simply connected it to exa, which is one of the best options available for all of the detailed research that to any API.
06:07
So I prefer Excel for that one.
06:09
then I am simply passing in the tools input arguments for my major input is I simply.
06:15
This variable refers to whatever the user is asking.
06:17
For example I can pass in any topic or any specific information that I need to get my research or on.
06:22
the next thing is a basic LLM door.
06:25
So for here I can simply.
06:28
Yeah, so this is just like to, to provide it user provide inputs and generate responses.
06:32
I am giving it one LLM node and in this one I am passing it.
06:37
so again I selected a model and I'm passing it a basic very simple prompt like Twitter LinkedIn post about whatever the information was or whatever the research was done through the last two agents.
06:47
and then this is going to give me the final caption for this agent source.
06:52
So what I can do is that once I have connected all of these nodes I can from this option I can validate the flow so it checks like all information is provided accurately, there are no issues and then I can search it.
07:02
I want to clear the history first and let's start with the new one.
07:05
So I would say maybe the topic should be AI agents for small businesses.
07:12
And now this was the topic.
07:13
It should start off the information from researching on this thing.
07:17
and after it gets the information from this tool about the detailed research it should then go on and create, create a capture out of So yeah, so I can even show you how what I got from the research, all of this information from the tool output it did.
07:32
Finally, what was the input or the, this.
07:36
So see how it has provided us all of the, you know, the detailed caption based on the research done and I'm also getting some links in it.
07:44
setting this.
07:44
This pretty much explains what how Agent Flow is powerful enough from any SQL chatbot and how we can create one.
07:51
and I'm definitely gonna share more use cases in the coming videos, videos, so stay tuned.
07:55
I think my final take on this would be that I like we all know that agents are incredibly, incredibly powerful and we can do tons of amazing stuff with these.
08:04
But honestly I would not just switch to agent flows just because they are in hype and they can do lots of stuff.
08:09
For my own use cases I would still prefer chat flows or simple chat.
08:13
for most of the stuff.
08:14
like if you are using like if any of your use cases are related to you know, simple text transformation, like the steps never change, you want speed plus low cost.
08:24
You just need an you know, output from a basic input.
08:27
You need sort of predictable behavior and your overall use case is just about you know, rewrite, sub reads and translations that it.
08:35
I think chatbots are good enough for it.
08:37
But when we are talking about agent flows, it all comes down to again your use case.
08:42
But what I feel is the best reason for using any of the agent flows would be where you need something like the AI must figure out the steps steps or you need research decisions or there's branching involved or you want most tool usage, you want to call different APIs, or there are lots of loops or revisions Then you need an agent it's not about replacing chat flows, it's about choosing the right architecture for the right problem and that's what you need to identify.