hi guys my name is Karthik and I from is automation.com and today in this video be talking about what are AI agents and how to use them for automation testing so before we get into this automation testing part of AI agent let's talk about what is this AI agent itself well AI agents can be defined in several ways uh and some Define this AI agent as a fully autonomous system that can operate independently for extended period of time while some other use the same term a a agent to describe it more in like a perspective
implementation that follows a predefined workflow so you set the workflow and that's guided workflow is what this AI agent is going to work with but some others say that this AI agent can become like fully autonomous and it can take action uh by them themselves so either way AI agent is going to enhance the effectiveness of this AI system which will enable the large language model to interact with the external world because we know that the large language model which we are using at the moment with the chat GPT or CLA or Google Gemini these
large language models are limited with the functionality to generate an output or maybe generate an image or video or whatsoever but they are not really interacting with the external World well they are not enabled to talk with the external world but this AI agent is going to make that happen which is pretty cool so essentially this AI agents are a computer program that no longer ER needs to function as a human control Tool uh which is confined for a specialization task rather they can now combine multiple tasks without having human Intervention which is which is
pretty dope because that is what is going to make this AI agent pretty amazing and that's what is really happening and there are few examples of this AI agent like the one which was released yesterday uh from the open AI the operator uh which is pretty much like an uh like a AI agent which is going to do quite a lot of things for you I don't really have an access yet it's only for the chat gbd Pro user with the dollar2 200 subscription limited to United States of America not for New Zealand people like
me so yeah that's the AI agent which the op just released so there is also perplexity which have also released another uh AI agent by themselves uh which is Al going to do pretty much like uh like what this openi is really doing so they have released an agent and there is already a lot of Agents which available uh like the hugging face small agent uh and the uh GPT uh the auto GPT agent that you are seeing over here and there is also this uh Cloud anthropic mCP uh server which also do the exact
same kind of thing I've always talked about the cloud anthropics mCP agent which can do quite a lot of different operation like model control protocol which can go and talk to the external world we also have a plugin for the mCP uh playright which is going to do pretty much like automation testing on the UI as well as for the API using that agent functionality so we talked about it we also have a GitHub reper which has got all the details as you can see over here so these are the things uh these are the
AI agents that we really have got over here but this AI agents level can also be categorized uh based on their capabilities as you can see over here so if the AG gentic level of the AI agent uh is like like there is nothing then the model has no impact on the program flow the developer has the full control of the prr and he will just do uh all the operation by it himself like if you just do print LM output then the LM response is going to be printed but if the agent level is
like one star then the model determines the basic control flow but still here the developer has the control over all the possible functionalities but still guess what the um the model will do like determine the basic flow like if condition switch case statement like that but if you have an agent level with like two stars like two level upgraded then the model determines how to function uh how the function needs to be executed so now the the AI agent will get the capability of doing it for you which is amazing right and then if the
the level of the agents just increases to like three stars then the model controls the iteration and program continuation and now you see that it is just becoming more and more smart and the developers control the high level function of a system and lowlevel systems will be taken care by the AI agents and once the agent level become like four star fully capable then the model writes and execute new codes by themselves and here the developer is no longer needed and the developer defines the high level function and the system can which the system can
do and the system can control all the possible functions uh and when they are done they will say that the task is completed so this is the four star that we're talking about this is exactly what we are going to be seeing as a demonstration over here with one of the agent called as hugging face small agent I've already talked about this small agent before but I'm going to show you bit more detail how we can use it for the automation testing purposes so let's quickly jump into that particular uh demonstration I will show you
how amazing this tool really does things so as you can see I have just created a very very super simple code uh with the small agent in the python over here and and this is the line that I have cre over here like I have imported the small agents uh Library uh and I have also imported the following class from this particular Library the small agents library and the code agent is the one which is responsible for uh holding all the different tools as well as the models so if you're going to have uh a
tool if you I mean for definently because we're going to talking about the code which is going to talk in the internet to grab some information you need to pass this tool like Duck Duck Go search Search tool this is going to perform the search operation on the internet for you that is the tool that we're passing in over here and you can pass any number of tools that you wanted I mean there are many different tools available you can pass it as well as an axillary help for this particular agent and then we are
going to pass the model over here with this hchf API model so the default model is going to be Qin model which is going to be chosen from the hugging face API uh with the hucking face Hub but if you want to specify any uh any model as well you can do it for instance you can pass GPT 40 if you wanted to or you can also pass Cloud anthropic and you need to supply the API key for that matter and once you do that it is going to talk with those agent to perform the
operation so the code agent is not limited to any specific model it's going to be uh it's just open for any different model that you have uh specified over there which is amazing and now in this particular agent I'm just going to say run and I'm going to specify some task which it needs to perform so I'm going to say uh perform or write a uh code or uh maybe I can just copy paste the one that I have already written over here for playwright instead of playright I'm just going to say selenium so I'm
saying write a selenium test code to run the test in an headful mode for this particular URL uh uh and then perform a login operation uh and then navigate to the employee list link and create a new user with some realistic data see I'm doing quite a lot of things over here I'm telling this agent to perform these operation for me so this is the command that I'm giving to the AI agent to perform these operation and the AI agent is going to do all these things for me so just to show you uh on
the context of what we're talking about uh with the AI agent so if I go to this application over here uh and if I go click this login you see that I'm going to get this particular link or URL and I'm asking the a agent to go and login for me without me supplying the username and password whatsoever and I'm asking the agent to go and click the employee list click this create new button and enter some realistic information on this particular uh particular window and make the task completed and it's all going to happen
on my browser it's like it's not going to be happening somewhere in the large language model or Cloud anywhere it's going to happen on my machine with the agent and that's when you are letting the large language model the capability to go and talk uh to the external world like my machine uh to perform any action that is the power of the agent itself guys now the agent has got the capability to go and talk to the external world with this agents over here so now I'm just going to save this code and let me
try running this code just three lines of code it's going to do all the Magics for you and see what are the Magics going to happen over here so the moment I say Python 3 small uh agent demo the small spelling is wrong it's just SM o uh agent uh not SM o l e agent or whatever you see that now it has chosen a model for me which is the Qin uh coder 32 billion uh instruct parameter model uh and now it's going to perform the action over here for me with selenium python code
and it is trying to run the code but before itun the code it is going to search on the internet like what are the website which has uh which it has got to perform the same kind of operation that we have told to perform that operation it is trying to learn all the information and now it has created a code which it can execute on my machine right and then it has written a code for it as well like how it's beautifully doing that and now it has really failed I mean this code is meant
to fail and I there is a reason why I wanted to show you this time is that you see that this selenium import has actually not happened and because we are telling the agent to Performance selenium operation the selenium does need some libraries to be executed the code needs to be executed on the context of this uh small uh agent uh box so how do we actually let that happen so we need to somehow tell this uh this agent to hey uh you know what import this web driver library or serum library for me just
let this uh thing to happen see because we have not let this happen to the agent now the Cod has just manually provided me the steps to perform the creation of a new user on the website and the test has got I mean the task has been completed so now I'm going to let the agent to really get that capability of executing me executing on my machine and the way I can do it is I'm going to set the additional authorized inputs uh over here uh I'm going to say this is going to be an
array sorry like that and I'm going to say uh selenium so go and have this selenium thingy uh for me and also s web driver common key uh and common uh buy uh and then probably uh let's say web driver manager. Chrome as well if you remember the web driver manager is also required uh if it's going to launch the browser so I'm going to give it just in case if it doesn't really require that it's not going to really use it so I'm going to give all these three Imports like an additional Imports for
the agent to perform the action for me and now if I try to run the same exact code you'll notice that it is going to perform all the action by having this import capability now hopefully it is going to the agent is going to open a browser for me so now see that it has got all the learnings from the website like how to automate an application and the step one it is now writing the code got it checked and now you see that it has opened the browser for me and it's entering the username
and password without me supplying the username and password which is crazy like how it has learned even my username and password from the uh internet uh and then it is trying to uh perform the action but see that first time it couldn't be able to achieve the uh operation successfully so it has just abruptly failed there to perform any action on the UI uh and and see that just with a log off like login operation it just failed now it's trying to do the same login operation second time and third time uh and now it
is clicking the employee list which is also happening and should create click the create new user it's done and now we should enter the user details so now the AI agent for the AI agent this this page itself is new right like it needs to learn that particular page the agent has now learned that page see now it's running third time to see if it it could a ble to perform any action on that particular page because it has learned the page this time you see that it's now learning the page uh with all the
uh page source as well behind the scene uh it is gaining all the information required to achieve the task and for the fourth and final time this time you will notice that it is going to go create the user with the information as you can see there which is amazing right like it has learned the page it has performed many tasks to make that happen and the action is completed which is amazing so you see that now the AI agent has reached like a maximum step but it has really achieved the goal of completing like
creating uh creating a user and it it has successfully created a new user on our website which is amazing right like this is what is the task that the AI agent is doing a lot of re TR operation to perform this operation and this is amazing already because now with just few key strokes from our side we could able to do an automated UI testing using selenium with this agent and I have tried even further to see what this agent is really capable of can it really do any API testing like how we do like
a big workflow of testing so I tried that as well so I have actually went a level further and I just went to um this particular website like uh the API uh restful API dodev like restful uh api. Dev website you can see that this is an API uh which is available 24 bar7 for free to perform our testing purpose so they have got some end points over here you can do uh like a get operation post operation put patch and delete operation you can do every single thing with this particular API endpoint and you
can also store the responses and do all things there so I'm going to use this particular uh apis that I have got and I'm going to do an test over there so I have already copied the uh really I have some commands that I wanted to perform the action over there so I'll show you what I really did this is the same kind of operation I tried doing with my mCP playright server to to show you how the mCP like Cloud mCP can actually use play's API capability to do all this action with the tool
that I Define fine but over here it's it's like next level where you don't have to do anything that agent is going to take care of that all of this for you so I'm going to say prompt is equal to and I'm going to just put like string literals here I'm going to paste that command over here essentially my prompt I'm saying is perform a post operation for me for this URL this this one with this body so I'm passing the body of that particular uh post operation and I'm also telling that verify if the
response is create has a created at and ID properties and store this particular ID uh in a variable like product ID which I'm going to be using as a future reference so basically I'm just telling like a like a postman where you just store like a variable uh in a for your operation so you can do that as well and then I'm going to ask the same agent to say like perform a get operation for this particular uh um endpoint with the product ID that you have stored in the variable right and now you also
perform a put operation for me for this same product ID by updating the value and then also perform a patch operation just to update a specific uh value out from it and also now verify if it has if the response has got an updated at property uh with this value you see that I'm telling a lot of workflow operation for API testing to the agent and now the agent is going to do all these things for me hopefully so let's see if that really works so I'm going to just save this code uh and I'm
going to comment this serium line of code and then I'm going to ask the agent to run the prompt for me this code is also going to fail for the first time I'll tell you why because we are going to be asking the agent to perform uh an actual HTP request operation so we have to uh tell the agent uh or maybe let the agent to give the capability to go and talk to the external world and you see that now the uh code has been written but it couldn't able to perform any action because
the requests library is missing uh from for the agent so essentially this is pretty cool because you see that now the agent has got like a simulation task to perform like a get request verified put patch verified but it's really not doing anything behind the scene it's just trying to simulate as if like the the operation is successfully done but it's not really doing anything on that actual API so now let's give the capability to the AI agent to perform this action so I'm going to go ask the agent uh to I mean go ask
the code agent to have this Library imported as well so I'm going to let the agent to go and import this Library uh for your operation and now if I'm going to go and try to run the same code that we executed before this time you will see that the agent the code agent has got the capability to go and talk to the external world you see that the first thing it did is like post operation and it is trying to oh yeah there we go it has did that and you see that it is
getting a response coming up over here and has got the ID over here and then it is going to I mean this is the get operation that is trying to do yeah there we go see that the get operation is done the post operation is also done uh and it has stored the particular ID if I'm not wrong uh and it is doing a patch operation it's also doing that successfully over here see that the patch operation is also working uh and finally it is going to perform the yeah and finally the patch operation is
the final operation right like so it has completed all the operation and it has successfully completed the those things like just 2.48 second it has completed all the API testing for us and the agent is now capable not just the UI testing but is also capable of doing API testing and as you can see which is mind-blowing that's it guys this is how we can see how we can use the power of AI agent for automation testing uh as well as uh the API testing in much much easier fashion with just like manual commands like
a user command with a plain English text and it's just going to work just let me know your thoughts about how these things are really coming through and how it's changing the landscape of testing we can probably discuss even more detail about the AI agents in our upcoming videos of this particular YouTube series but I'm sure this is quite exciting once again thank you so much for watching this video and you guys have a great day
