16 Months of Building AI Agents in 60 Minutes

hello Builders so this is going to be a special video all right slightly different to the ones that I'm used to doing in this channel because in this video I'm going to go through the 16 months that I spent building AI systems for myself for my agency and also for my software company for my AI software company and I'm going to be discussing tips and tricks fundamental concepts and everything between that you need to understand in in order to build AI systems that are valuable for yourself and also for your clients right so let's get

straight to it so I think the wise place to start is to show you uh some of the stuff that I buil for FG Labs so I've built conversational chat Bots uh well conversational agents in bpress conversational agents in voice flow as well here you can see a complex workflow that I built for one of my clients here you can see a simpler one here we have a pretty big one and believe it or not this is onethird of the whole workflow and this was an appointment setter I built for a real estate agency I've

also built voice agents on Bland Ai and on vapy as you see here this is an example of an exist of an assistant that has access to three tools on make to book calendars get the current dat in time and so on and so on and whilst I was building solutions for my agency I've also built on my AI software company and the goal with this company is to bridge the gap between data and insights So currently we are at the demo phase we are getting on demo calls with people that would be interested in

the product and then you know getting their getting their feedback in order to improve the product and yeah so the reason I told you that is because I really believe that from my experience on the past 16 months and from the including the things that I've shown you that that I've built or that I I am currently building on I think I can definitely provide you some value that you can then take away to build uh useful stuff right and throughout this video we're going to answer this question how can how can AI agents help

businesses all right because this is not a simple question that I can just give you within a bu a few bullet points right to to to be able to answer this question properly we have to go through some Core Concepts that are used to build systems right using using language models so we're going to go through that and I think the best way to start is by introducing the basics okay because we're always discussing all the complex things that we can build all the all the all the you know amazing tools once they're finished and

so on but I think it's really important to go back to the basics all right being able to understand understand the basics would allow you to then build on those Concepts and then whenever you have a complex system that fails in some way you will be able to understand why it fails okay so so yeah understanding the nature of language models and how they fall into these larger systems will definitely help you with um problem solving right in the case that um you know you actually build have to get a problem from a user or

or for a client and then build this a solution out of it or troubleshooting an existing system that you have right so given the situation you need to understand the basics right and we're going to go to the basics of the basics okay because some of it might not be um might not be that straightforward okay because you get thrown with many different terms online you know agents assistants and so on and so on so let's go through it step byep step and we'll we'll build it as we go all right so what are large

language models well these are programs that have human language as an input and then we generate the text output okay as simple as that but what's what's what make it what's what makes these large language models so powerful well sure they understand human language but they understand the semantic meaning behind human language right what do I mean by by that well if I tell you that the movie was great what does that mean right on a scale 1 to 10 where where would you put the word great right so depending if I answer that question

or if my friend answered that question we're going to put that on a different scale right in in that 1 to 10 scale on a different part of the scale right and that's what semantic meaning is Right understanding kind of the abstract Concepts behind the human language right and this is what machines struggled before right before language models existed machines struggled with understanding these Concepts and now with language models here we are able to build amazing things because we're able to understand the semantic meaning in the human language right and then we have examples like

chpt googleart Claude that are able to do this um very nicely and here's are some ways in which we use language models right Q&A summarizing text generating code because code is also text right and yeah as I said before well well not before language models but a few years ago it was really hard to build and use language models and integrate them with other applications right we had to use libraries like tensorflow and pytorch and pretty much hardcode uh most of it and that's when line chain came and it made life easier for all of

us right this is an open source framework built on JavaScript and python for now and what it allows us to do is that it allows us to chain language models together with tools and actions right think about it as as a Cascade of of actions right it goes from one point and finishes on another point and thanks to that we're able to add or change language models together to be able to perform uh tasks of more complexity right and essentially what that allows us to do is to build applications that resemble real life things right

stuff that happens in real life in businesses okay and yeah it has to reduce the barriers to do this so gopl line chain definitely and just for you to understand what line chain where line chain lies in all of this you have to understand the different levels of abstraction there are right the different ways in which we can communicate to machines so down here kind of the most fundamental level we have CPUs gpus and storage and then we're able uh CPUs and all that CPUs gpus and storage are able to communicate with between each other

using an operating system right and then some smart people decided that well we can actually communicate within with the operator system using binary code as well another uh um another types of code but they all follow in the same level of abstraction and then some other smart people came by and they said well can we resemble this to the English language so that it's easier to um program and code all this stuff and communicate with between machines in an easier way and that's how we came with all these programming languages right and then some other

people said well I'm using the same 4050 lines of programming language to build something from a specific application let's say for data analysis why don't I bundle these 40 50 lines of code that are going to be the same every time I build a similar application and add them to a library right so now I don't have to write these 4050 lines of code again right so that that's what a library is and a framework is essentially something similar but for building applications right and that's what Lang chain is it allows us to build AI

powered applications quicker right so essentially this allows us to program more effectively and more efficiently and in less time and then we have these other abstraction layers and then we get to the application Level uh level okay or layer and this is where most most of us are when we build chatbots agents automations and so on right make is an application n is an application Instagram Facebook YouTube right and and my software as well so I think it's good it's good for you to just consider what all these are especially where Lang chain lies and

how it helps programmers right and where make N1 and all the applications that you use standby right it's in the application Level and funnily enough make anent on all these applications that have ai modules in them most of them are probably using line chain in the background all right so I think this is this is pretty pretty interesting and somehow useful to to consider right but anyways what can you build with language models using Lang chain well you've probably built stuff with na10 or make or or zapier for example right so this shouldn't be um

very mysterious at the moment but yeah conversational chatbots agents systems basically anything that involves language model inputs I mean language inputs sorry now let's touch on the Lang chain Basics so the The Core Concepts that um you can build applications with using Lang chain and the reason I'm explaining this is because I want you to have an unbiased way of calling all of these different parts of an AI application or AI system right because in YouTube or wherever you are when you hear about Ai and people talking about AI you get thrown with different concepts

and words right assistants and so on and this and that so I want you to have an unbiased way of calling all these things right and what is a better unbiased way of calling things than going to the actual framework that all of these things are built on right so we're going to go through these Concepts I think I just laid out two so it's not too many so we have the concept of chains and chains are basically just um language models and tools just placed one in front of another and they basically all work

together to achieve a result okay especially the language models so the the concept of change is specific to the language model so the the thing to understand here is that you can make a task a complex task that involves language models chained together so that the input of a language model is an output of a previous language model right and that way you're able to perform complex tasks okay and you see some examples here but don't worry you don't need to really focus on that you just need to you just need to understand the concept

for now and here I have a resource that I will put in um once I share this I will share this presentation and you will be able to click this okay and this takes you to a mini that um these people have have prepared so that you can understand this further if you want but yeah so as I said you probably have seen this before on make so I'm going to show an example really quickly and this is an example of a bland AI component of a bigger system and what this does essentially is it

gets a transcript we pass the transcript onto this language model and then this language model is going to say well is it qualified or not so it's going to spit out a yes or no and that yes or no is going to be used in this language model and create an email right so if it's a yes it's going to create one email if it's no it's going to create another type of email and that's the concept of chaining right so chaining multiple language models together to achieve a more complex task but remember these are

in Cascade so it goes from one way from for starts on one point and it finishes on on another one right so it has a flow it has a predictable flow to it and now we have agents okay so let's clarify what an agent is once and for all so an agent is a language model that has tools right but that's not all this language model with tools can make decisions so it acts as reasoning engines it can reason what tool to use based on the query it received and the environment it is in right

if you define it and that's what makes it an agent it can reason what tools to use on a given situation right so as I said it can interact with tools it's autonomous so it doesn't need a constant human input as you would have with a normal or a basic conversational chat bot where you just chat to it it's giving to you an output back right it's a bit more it's a bit more than that you give it a task and it will achieve it without annoying you back with with further inputs that it needs

right and as a consequence of that some agents are able to autocorrect themselves if any errors appear right so inside the agents term there are different types of Agents all right we'll see that on na10 because I I prepared a little tutorial or a little demo that we can go through and um over here you you will be able to understand uh what types of agents are there right and now we have an example if you know that that wasn't very clear we're going to go through an example of on what makes an agent an

agent right so the person is going to come and going to say well going to show me a graph of the performance of Apple stock from the last six months okay so we have a couple things going on here I wanted to show me a graph and I wanted I want it to be of app St stock from the last 6 months so there's there's a bit of complexity there right now that a conversational chat will will not able to do because first of all it doesn't have access to tools and second of all it

can't reason and and choose the required tools so first what's going to happen is that it's going to check for the current date and time okay this is something that most solutions that you guys are building need anyways because all these language models have a cut off date right um at least for the latest ones it's like I think it's 2021 or 2022 I think but yeah it always should check for the current dat in time since it's not a an inherent feature these language models have right they can only access without any external connections

they can only access their training data and that has a cut of date so they're going to access the web API get the current data and time and then he's going to say well today is the 7th of September so six months back mean that means that we're going to search from March to September okay so it does search all that right and then it searches for that data from appple stock um in this in this previous um uh in this API and then once it has that it's going to use that data and generate

a graph right so it's a multi-step thing so it's able to achieve all of this with without further human interaction right we tell it once and then it's going to perform all these tasks autonomously and that's what makes agents so powerful they basically act like humans in some way right and now we get to Rack systems so rack stands for uh retrieval AED generation and it's basically a technique to counteract a limitation that language models use to have okay and this was the fact that language models could not access uh didn't have access to knowledge

bases outside of their training data okay so whenever you ask a language model that's not connected to anything outside of itself you probably get got something like as of my training data I can't give you an answer all I know is X right so attaching a rack system on it would allow the language model to have knowledge on a specific uh domain right or a specific set of files that you give it so we're storing this information in the vector store and then we retrieve the relevant info to the user it's going to become clearer

now okay don't worry and it combines Three core elements all right the embedding model the vector store and the and the actual language model okay not an agent a language model and you probably have seen rack already so whenever you upload a file on voice flow or bpress you have you have actually used rag right you have you have uploaded your file your text file or or your PDF or your image into some sort of vector store right so so there you go so that's where you have seen rag models or rag systems and this

is a rag system with which we can do question answering on top of a series of files or data that we first upload to a vector database okay so this these files or texts get turned into vectors similar to how we're doing here right an embedding model turns text into vectors and then those vectors get uploaded onto this database right and they look something like this okay all all the words in those files are going to be clustered together right by similarity same as happens here so this is the animal cluster and this is the

fruit cluster for example um so that when we ask a question going ask that question turn it into a vector and now that um that those vectors that represent the question can be placed on this graph right or chart and now what we're going to do is we're going to pick the the nearby neighbors right of those vectors that represent the question we're going to pick those and we we are going to push them back to the language model and the language model is going to process that and gives us an answer right what we

have essentially done there is just pick the most similar words associated to the question right and we retrieving it back hopefully that makes sense but if not you can always um you can always check some resources that that explain in depth how how this works right but but yeah let's let's let's continue h and yeah you can you can read this but I think I I've explained it right and the point with this um also that I wanted to make is that rack does not equal AI agents okay rack systems are designed to only enhance

uh the language model with additional knowledge right but they're not autonomous right they don't solve complex problems their goal the goal of a rack system is to retrieve additional information from a database all right and enhance the language models knowledge if you want the this to be part of an agentic framework or part of an agent what you would do is remove this poor guy you would add a language model perhaps with other tools and you would turn this whole rag uh system into another tool all right now we have have an agent right so

what I'm essentially uh explaining here is that rack can be part of uh an agent's tool okay so that's my whole point on this all right and now that you understand these Concepts including AI agents and rag now I can present you to the concept of multi- agentic systems which you've probably heard of before but if you haven't well this is basically the AI agents that I told you before that are able to think for themselves able to use tools right and they're able to perform relatively complex tasks but now we're putting them together in

a group so that they're able to chat to each other and perform much complex much more complex tasks together right and this is this is definitely I think the path that we're heading towards in the future right multi- agentic systems because essentially what you could ask to this multi- agentic system is to create me a game from scratch right or create me a software an entire software by scratch right so you would have a an AI agent that is a programmer another AI agent that takes care of the product another agent that takes care of

the marketing another one that takes care of the infrastructure for example right and they all work collaboratively together to build this app or software right and I think in the future this is going to replace play a lot of teams so just imagine how important this is that Microsoft decided to come up with autogen right so this is how crazy it is then we have others like chat Dev and then you know with stag you're also able to build um um multi- agentic systems right bit more visually which which I like I love what they're

doing I tried it and and I think it's great but but yeah it's a bit expensive so it really depends how much budget you have you could go with this but yeah these two require code right but anyways you get the idea of this and one important consideration that I realized when Building multi- agentic Systems with my team was that if you have a large language model and you tell it to do three different tasks right let's say you give it the ability to do three different tasks that agent is going to do much worse

at doing one of those tasks compared to a smaller agent or language model that is only able to do one task right let me explain you what I mean in a in a drawing or or or an illustration okay so we have a large language model here okay and let's give it the ability to be able to write me an email write me a Blog and also write me some specific code okay some I don't know what type of code but some specific code and then we have a smaller language model that is able to

only write me an email okay this larger one is going to be much worse than writing an email compared to this one right which is going to write it much better and the best way to think think about this is think about agents or language models as humans all right if you train one human to do either emails or blogs or codes it's going to do much worse at writing an email that than someone who's only been trained in that span of time to only write emails right does that make sense so that's that's something

that I realized and what you can do then if you understand that this is the case then what you can essentially do is let's see if I can get get it here you you can have three smaller specialized models right that each do the different tasks that the bigger language model can do and then you're going to find out that all of these specialized smaller models are able to do the these tasks much better than this bigger one right which is which is really nice to know and you know in order if you want to

if in order if you want to ask um if you want to ask the the this this framework to do either of them that what you can do is you can connect these to another language model or an agent right this case you can call it agent to decide what model to call right so this is a really good approach to building robust applications right given that the tasks that you're placing here are not super related to each other right because if they are super related and you do need to retain the context because uh

from here to here you might lose some context right whereas in with with a within a language model within a single language model you're not using any context because you're not passing any messages from one language model to the other right but if that's not the case if the tasks are mildly different then you can use this approach to get better results right um I don't know exactly where you're going to apply that if if you program then that makes sense and if you build multi- agentic systems um that make more sense well you can

actually um build these on na10 or stack AI actually you can choose probably the the the model size if I remember correctly but but yeah that's a very important takeaway that that you should uh take with you when building and designing multi- agentic systems right the size of the model and and the type of tasks that they're doing and see if you can uh um you can you can organize them in this way right and you find out that they they they can they can end up doing um they can end up performing much better

right and obviously this depends on the depend of each task right and how much context do you have to have from one task in order to do the other one right this all this all really depends but that's something for you to try definitely if you are building these type of systems and this is a big one right I wanted to clarify this so AI agents and AI automations are different things right you can you can you can make AI automations act as AI agents right but the way it's portrayed out there the way it's

actually done and and shown on YouTube videos is that AI automations are simply kind of systems that do specific and repetitive tasks right they reach from point A to point B okay so similar to what's happening here right I'm going from point A to point B there's no there's no part in the system with which in which we allow one language model to just perform multiple tasks right you can essentially tell the language model to to spit out a yes or a no or a maybe and then have three different routes that if the language

model says yes well it's going to do that if it says no it's going to do that right but that's embedded within the ultim right it's not part of the language model if that makes sense the language model it's it's it's kind of different okay because that's built within make for example those filters so it's not the actual agent triggering them if that makes sense but but yeah so it does it reacts based on predefined workflows right we are defining all of that but AI agents can make decisions and take actions independently and they do

adapt to new situations right and most importantly they also learn from interactions right which is what these completion modules cannot do these are input output but they don't learn they don't get better as they go right they get better if you write a better prompt but that's about it right so hope that was clear and once again well why are agents a big deal I'm I'm probably repeating myself a but they can handle complex human tasks right and the second point is kind of the same but yeah the speed of processes that require manual effort

I mean this also happens with AI automations and yeah they enable humans to focus on creative and strategic roles I guess they this is not really um unique to agents but definitely agents can help in many places that AI automations cannot right I'm not telling you that AI automations are worse that agents I'm me telling you that they are a bit more advanced right and if you combine these agents into groups that's where the real power lies as I explained before with the multi- agentic systems right but anyways you can you can skip this this

slide but anyways I think it's time to build your first agent okay and remember a agent is not not doesn't necessarily mean it's an AI automation okay it's for you to know and you might think well you know do you need python use IDs no frontend host on servers Etc well you don't need to because we have tools like NN right and you might think well what what's so good about n10 well the good thing about na10 is that the way it's built it has uh an AI agent um set of modules that are going

to show you how L chain like you know AI agent components in L chain work which is very very helpful make doesn't not do that sabier does not do that right they're too abstract they're too abstract for you to be able to see the components of an AI agent right but this does right I'm I'm going to show you and the good thing about this is that all all the all of the logic can be handled inside any1 so inside at Ann you can build the AI agents as well as as well as um triggering

some automations and workflows right whereas with these tools you would have to possibly connect voice flow with MC right and that would technically be an agent right but you have to you have to connect two platforms together right voice flow WordPress or you know what I mean right so getting the conversational AI chat Bots together with AI automations okay voice agents they're definitely agents right because they have they have the the the decision capabilities a bit better defined but but yeah but with na10 you can definitely see the different components that make a real AI

agent and now I'm going to show you an nbnb chat board that I've built on n10 okay so let's go ahead and look at that all right so regarding the AI agent components and so on this is this is what I explain you about right so this AA agent and then you're connecting the chat model then you're connecting the memory and you're connecting the different tools that it can use right the vector store tool the pyth con Vector stores right so this is this is essentially the the the rag I I showed you well part

of the rag system that we're showing you and you see here it's being used as a tool right what's going on here is that we are using this agent and we are retrieving important information whenever the agent needs to to answer questions about the Airbnb okay so I have something here on the background right on another um on another workflow on na10 and what happens is that um right now it's just it's just on test right so I need to click this for it to work but what it's essentially doing is it's going on my

Google Drive and downloading a file onto a vector store right and then it's using an embedding as I told you before uh in the rack section and and it's yeah and then it's using all of this to store convert that convert that information into a vector and storing it into this Vector store right and once that's there then we we we're able to use this as a tool okay so let's let's chat to it okay so let's see what it says so hi right so what I gave it was the following right you might remember

this from from another video right so airbm policies so I'm going to ask it the checkout time and I'm going to ask about the house rules okay let's see if it knows about it so let's go back here oh time should I leave the ABNB for checkout right so you should plan to check out other BB before 11:00 a.m. local time right and the cool thing is that you see on the right here you're able to see all the steps that are happening right so for example here I'm asking it what is the check checkout

time at rbnb and we're seeing here how it's converting that into vectors right so we able to see this visually so the level of control that this has is really close to what you would have if you're programming this using an editor right if visual studio for example using python right and this is probably the closest that it gets to that so that's why I'm saying it's abstract but not too abstract right like make or zappier where you know it gets too abstract that you can't even see this stuff right you can't even see the

embeddings working in the background you can't see a lot of that so that's that's the power of this and then you can see the actual system messages and the stuff it's it's it's being uh added on on the model and so on so you see the buffer memory working and and yeah so you're able to see everything which is which is great right let's ask something else so let's ask let's ask what are the house rules right so here are the house rules for theb property no smoking no parties and so on right and you

can imagine how you can connect these to tasks these uh to other tools right maybe you want to connect it to let's say um search API similar to how you do a make for example and let's see what else you have Wikipedia as well right so we can trigger that who is I don't know if I spell correctly that well let's let's say who's Michael Jackson right right and now you see that I use Wikipedia right so we quer we quered Wikipedia and then we retrieving all of this back from Wikipedia but then we are

process it a a bit more so that we get a um a better answer right a condense answer so there you go right we just Ed the Wikipedia tool and we can see that it's being used now we can possibly use the calculator these are really simple um these are really simple tools right so the calculator but you could imagine you can you can attach other tools right so so let's see what tools it has well you can for example look at this look at look at all the all the things that you can connect

to right it's pretty much so the great thing about n10 is that probably I mentioned it before but I'm going to mention it again is that you can use AI agents you can build AI agents as you would in code mode right because you can see you have full control all of the different components but you can also use these AI agents within an automation framework right like like make for example so you know you can you can add you can add different actions to do right you can transform data and so on so that's

that's the power of na10 right and you can build really complex systems with this without having to rely on two apps at the same time or three apps at the same time okay there's of course the different pros and cons with an10 right for example because it's an AI agent you wouldn't have the flexibility that you would have um when you build a conversational agent in bpress for example where you exactly understand how the conversation is going to go but but yeah this PR and conso it really depends on your use case right but if

you really don't mind about controlling too much of the conversation then you can definitely have something like this and it's going to help a lot and the way you set out the pine conector store right and the way you upload files at least how I did it here is that you obviously need to go to Pine con right this is where we would create our indexes inside a vector store right and a nice way for you to upload this is by creating a separate workflow that I showed you right you will go and do some

authentication right like you would do with make for example you choose file download you choose the file itself you can also choose folders right so that let's say you keep uploading files to that folder you can automatically trigger this so that it gets uploaded automatically to Pine Cone right so obviously it takes some more time and expertise and kind of knowledge to be able to build these Solutions but the good thing with this is that if you're looking for that flexibility that make and zappier doesn't allow you to have then you can build here right

that's all it is and then I have here WhatsApp um so yeah I I have a WhatsApp chat board as well but I don't want to get into the the details on how to set this up I was thinking to make a separate video on that if you want to if you want to see how to build a WhatsApp chat bot on any 10 please let me know down in the comments below I'll be really happy to do that but I need to see if if people would be interested in that right but but yeah

it's it's really cool app as I said well I I didn't try try for long right but but it seem it seems great and it definitely has a lot of capabilities on it so yeah so yeah that's the demo and I listed the different Ai No code tools that you can use right and their limitations so this is a conversational AI chatbot that can be turned to an agent if we send Json apis uh Json apis if we send um um we make API calls to other software such as make or zappier or to another

server that can perform a task over there right but natively it's not able to do a lot of um a lot of tasks by itself right so that's why I was saying that you know uh n10 basically is voice flow and make combined in some way right but yeah voice agents are able to do that they have that autonomy to just um trigger tasks easily or easier but yeah but anyways I want you to be aware of the adoption of agents in Enterprise and AI systems the ly JB Morgan Len all the big companies are

doing that already it's like a race to see which companies actually integrate these faster into production right but the only thing I want to point out here is that we are here we are at a really low market cap at the moment right with with the adoption of agents in the market so we have this kind of parabolic curve that we still haven't touched on yet so just to let you know you're still early so as long as you understand what I've mentioned in this video about air agents you're doing already really well right you're

you're in the definitely in the top 10 20% of people just by knowing that these things exist so so yeah so it's good to know it's good to know it's good knowledge understanding agents and and the capabilities because you can capitalize on it right if you're early on the market you can capitalize on that a straight away by building tools on any1 or as I said connecting stuff on voice flow and make and building it to however you want but as long as you act on it right now you're in time and you're up for

a big um chunk of the market that's for sure still early and now I want to touch on some considerations when building um AI solutions that are high quality okay first of all data is King right if you want to have an AI system that gives quality outputs you have to train it on the right data right you have to give it good examples right the the the the data that is maybe maybe it's retrieving data from somewhere to be able to generate a query or or an answer that data has to be high quality

right it all lies in the data that you feed the language model is if it's either for training or as an example or if it's using that to process um some other parts of a task right data is King and then you have to worry about PR jering because one of the reasons why your agent is not going to perform properly is because of the prompt you give it right super important to know how to prompt these language models and agents and this is a framework that you can follow as the system prompt for your

language model okay roll objective tools instruct instuctions right about the tools they have to be detailed and concise but that pretty much applies to everything right detailed but concise don't fluff because when you fluff the language model is not going to perform as good the way you want the stuff to be outputed and formatted examples very important to give it examples and then additional notes this is specifically used for tweaking the language model's output to how you want it okay okay I'm going to show you now a prompt here that I have for a clinic

so you have here the system role okay then you have the objective right what's the objective of the language model or the agent then you have the tools that it's allowed to use and then the instructions right what what what it can what can it do right with the with the with the tools that you give it and you see here that I'm not being fluffy right as for the col name inquire use the provider busy times but you see also I'm giving it enough detail for it to act right so you got to find

that balance between detail and conciseness right and then I have another instruction for another task output requirements examples right and then a little reminder a set of reminders at the end so this can be used in order to tweak it so let's say you you you try the agent you test the agent's output a couple times and you see something is not working well the reminding section is for that right if it forget something you can just let it let it know right right here so that you don't have to really change the main structure

of the prompt if you don't want to right you can always stick it at the end so those are little things that I learned where regarding prompt engineering which I think is as I said one of the most important parts when building a solution and some other useful things to know are how to integrate with other applications through apis and web hooks for example right so the first method to connect to to the first main method to connect to um a software or an application is through an API call right you need to be aware

of the two main methods that you're going to use get or post get is to retrieve something from a server some information and post is to post to send the information to that server right if you wanted to process something right when I say software I mean or app I mean a service right a service that you're using to perform or achieve an action right for example um I'm sending data to data for SEO I wanted to scrape stuff for me and then I want it to retrieve back some information right as the top five

searches or I'm sending data as inputs on a post request to apy it's going to do some web scraping for me and then I'm retrieving that information back right so first we're doing a post request right and then that service is going to perform all the things that it needs to do and then it's going to retrieve me something right which you could add a get request at the end to retrieve that data if it doesn't do it automatically okay then obviously you need to know the URL of that service so where am I sending

the information to and the request body what information am I sending very important so this request body is going to change depending on the app or the service that you are um that you're using right so you need to get well familiar with these request bodies because they vary depending on the service that you're using right then you have the request headers this is the authorization right so in order to send a message most of the times you cannot send a message without authorization without a password or a username because the service does not know

who you are right so obviously it's not going to allow you to use that service unless it's free if it's free maybe it probably doesn't need authorization but normally most Services already for security reasons uh need authorizations so that you don't have Bots just using the surveys a million times and then breaking the service rers right and then you have a response body that's if you are doing a get request right mostly but yeah study API calls they're really important okay you're going to use them nevertheless and then the easier method is web hooks right

you literally just have a web hook URL you're sending something and then you're expecting something back right it works slightly different but it's easier to connect to because you don't need to understand much about um request bodies and how the service is is um how the service request bodies are are are shaped right you could essentially send a pretty General request body without any specific structure and the web hook will receive it right we receive that information without any um without any red flags or anything right so here I have a web hook for example

right so in vapi let's see in vapi on functions it's tools Now isn't it it changed but on on on these I'm attaching a uh let's see if I can use it here um does it say here it's somewhere let me find it um here you see so this is a way to send um this is a web hook right so you see it says hook so I'm sending something to this URL right and I believe and if I'm not mistaken uh I'm sending the I'm I'm triggering let's see I can so what's happening here

is that the agent is is the agent depending on what I ask it to do right in the call um it's going to it's going to choose one of these right one of these tasks right book calender V get available slots and all of these are linked to a separate um automation right and each of the automations have a individual web hook a unique web hook right so what happens is that whenever I trck at the first one right I added here a web hook that is going to trigger that's that that's going to send

um some data to that web hook right and that web Hook is coincidentally the address of this one right it's the web Hook is this web hook address right sorry just to clarify so the agent is going to if if it needs to get the current data in time it's going to trigger this tool and this tool has a there you go has a web hook address attached to it right so it's going to send uh a request to this address and that address matches this address so what happens is that I'm going to trigger

that I'm going to trigger this Automation and I'm going to receive also a request body so that means that I'm going to trigger this but as well I'm going to receive some uh some um some Json object that I can possibly then refer to and maybe add on this language mod model or add on this Google sheet for example right so web hooks are a more General Universal way of sending data between uh applications and services okay and yeah if you can use this method I suggest it because it's much easier right unless you need

something very specific uh or something that you know oh this is more for calling services right this is more uh for calling uh servers like web hook or automation automation platforms right this is for data foro for um right as I said Services right stuff that that that is behind some software or something and you get the idea right as I said data Rao um what is the other one apy and so on so on right you're going to usually use these more and yeah so well that's this is what I was uh referring to

anyways and it's very important I think in order to understand these things correctly and in order to be able to know um how to manipulate the data effectively because there's going to be some cases where you're going to deal with um data manipulation inside these automation platforms I think it's good to understand how that data looks like and what it means to have the data in a certain way right and above that I think learning Python and JavaScript as well gives you the ability to think about problems in just a programmer way right if this

than that so you know thinking about uh specific conditional logic so if this happens this then I'm going to trigger this API service or I'm going to do I'm going to trigger this HTTP request and so on right so it gets you into that mindset of solving problems in a programmatic way right all that makes sense and yeah these are the main things that I think you should learn right variables and data types statements if else list and arrays very important to manipulate data and just knowing what person parsing data is okay authentication types as

well really depends on what you're doing but I really do suggest uh that you learn this because you're going to have to authenticate with stuff like Google which is a pain in the ass to authenticate with and and yeah in order to use some some of them Services you need to learn how to uh pass the access token or refresh token effectively um I think you need to do this as well with WhatsApp verification and connected with WhatsApp unless you're using something like um man chat which is slightly easier to use but yeah depends on

your use case right and another top thing is understanding the client's needs and objectives effectively right I think learning how to program in Python and and JavaScript is going to help you do this right you're going to be able to turn what a client needs into kind of a set of tasks that you need to do but honestly this comes with practice right and when I'm talking about practice I mean creating architectures right creating flows so let me show you what I mean for example this right so you need to be able to learn how

to build these it's not really difficult right there's a lot of tutorials out there that show you how to do it if you want to teach you how to how to build these let me know down in the comments below and I will help you out right and I will make a video about it but I'm not building anything I'm not I'm not drawing anything crazy right I'm just building a flow on how the conversation of how the tasks are going to happen right within within a flow and how the user is going to interact

with with all this within all this right I'm collecting um if I'm collecting any any variables from any of the options right um let's say if I'm connecting to the database or I'm turning um a file or a variable into some format or I need to be aware of a format I'm creating a new variable right stuff like that so this is so you just need to get into the habit of of of creating these because it's just going to make your life so easy because if you have these architectures on the side as you

build then you're not going to have to think too much about how to build it right it's already pre-thought you already thought about it in the background and you just have to follow the blueprint right but as I said you will get better ideas as you practice them okay and yeah just a final statement agents are the future because of all what I said before they're so powerful they're able to think by themselves they're able to trigger tools and then when you bundle all these agents together into a multi- agentic architecture well you you you

will be able to replace teams right I just want you to understand the importance of that and the importance on where we are at right now on the adop on the adoption of these agents in the market right so since they future they're the future I think you should definitely act now now that there's time and there's people trying to catch up to these Concepts right and yeah as I said they will handle monotonous specific tasks T that connect to software and use text voice Solutions already right this is happening with Bland and vapi and

what's going to happen is that we're going to see a big shift of on uh on on how humans work we're going to be pushed onto more abstract tasks right that agents will not be able to do like personal relationships meetings and Deals right more creative arts and so on and so on and that's it thank you for listening and yeah let me know what you want me to to do next right any specific videos on to on what to build on what to build agents AI automations voice agents anything that you want me to

do please drop me a comment down below and I will answer and I will consider it for a future video okay thank you so much and I'll see you in a bit