🤖 AI Agents Explained: Effortless Automation Testing Starts Here 🚀

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