AI Agents DOES THE CEO JOB!!! (ChatDev Tutorial)

Key Takeaways

to be honest chat Dev is the first AI agent that actually produced complete code for me so chat Dev is a communicative agents for software development it's like running a software company you've got CEO you've got CFO you've got CPO Chief product officers code testers code reviewers software developers and in this video you're going to learn every single step on how to run chat Dev on your local computer as a disclaimer you would need open a key that's something that you need to keep in mind the first step is for us to clone chat div repository once you have successfully cloned the repository then you have to enter the repository like the enter the chat Dev folder and then start creating a virtual environment before you create a I mean virtual environment is not mandatory but at least it's good for you to keep your environment intact check into the folder you will understand how the folder structure is this is just to make sure first of all you have cloned it properly but also to understand how the folder structure looks like currently there are certain important folders but that will get into later for the first thing that as I said you need to create a virtual environment before you install the requirements or txt let's create a virtual environment you can either create a conda environment or you can create a virtual environment whatever you want to do you can basically do that and once you have that ready then the next thing that you need to do is you need to start creating or start installing the requirements ready clone the chat Dev GitHub repository then enter into the folder create the virtual environment activate the virtual environment and the next step is for to install the requirements.txt it has got a lot of dependencies so make sure that you know you don't mess up with your environment so install the requirements.txt once you start installing requirement.txt with Pip 3 Dash R requirements.txt then you are ready to go to the next stage this point we have installed all the libraries and also make sure you have the open AI key added as an environment variable for this particular session so so then once you have done that now it's very simple and straightforward to run it use run dot pi and then specify the task in this particular case build up a timer and then you need to just simply say what is the name of the project in this case one LC and then you would start seeing the conversation between Chief product officer and everybody else like it's like a very long conversation it goes on to the details once the open a API is successfully hit then you would start seeing how many tokens have been used and all the details around it and you can also see in real time what kind of conversation they have what is the stage of the product and right now you are seeing this everything on Terminal but very quickly I'm going to show you how to do the same with a very nice graphical user interface where you can see the different discussions happening between different agents for example now you can see the model that have been selected and what is that CTO saying what is the software tester saying what is the programmer saying but like I said like there is a very nice beautiful way to visualize this let's understand how to do that the way you can visualize it is very simple there is a folder called online underscore log and that requires flask as an instantly requirement from that online underscore log just run the app.pi file and you would actually see this interface that has two components one is the chat chain visualizer the second one is live discussion that is happening between all the agents that you have got so right now as you know that we have successfully built the promoter timer because that's why you have got the user manual in itself you can also realize that the project has been completed by going to the terminal because the process is done then the next step is for you to run the software enter into the varos and then see the name of the project that you gave and if you have run it multiple times then check the latest time stamp so when you go to the latest timestamp you can actually see the project in and itself and once you enter into the folder you would see what is a file with which you have to run main.pi is the for the python project but otherwise you can also check the manual that actually the software team has created the agents the chat div once you run Python 3 main dot Pi because in this particular case it is a python project then the python file would be run and then you will get to see the python interface like for example if it is if it is a simple python CLI then you would see the CLI if it is something else you would see something else in this particular case it's a GUI the graphical user interface which is a simple promoter or a timer and I can start the timer and it stops no errors no issues literally we wanted a simple promoter timer and if we have literally got a simple promoter timer in Python that just simply Works no no strings attached just simply works even with the GPT 3.5 turbo APA I don't have gpt4 access and I wanted to check this and it works completely absolutely fine let's try to build one more project now this time let's try to build a very simple hangman but also monitor what the chat devs are talking so we are going to see the live reply of what they are doing our live replay of what they're doing so what we basically want is we want a simple hangman application as you can see here the city CEO is talking about building a very simple hangman and I did not specify what what I want like in hangman it's I just said simple so at the end of the project you would notice that it is going to be very simple intact so this is the first message starting from the CEO and you know you can see the ctu's response who is going to decide on the technology and then you would see this conversation going back and forth with the different role playing happening so they've decided at this point python is a programming language in which we are going to do and they're also very quite very strict or very diligent about documenting everything and also now they what I mean they here is this AI agents not the chart Dev developers so these AI agents are quite diligent about creating the right document and the go component by component as you can see now they're talking about the main file that's what they're going to start with but then they are going to go ahead with the different sections of this particular game in and itself where they will say that okay now main is finished I have to do this I have to do this I have to do this and you can scroll through this chat replay to understand what is that component that they're building now and what are the components that they should start building and before ending they're going to start giving it to the chat tester or the software tester who would actually test all these things as you can see they started building these classes which is a strong object-oriented culture and with the object-oriented culture they are also identifying what is the role like right now programmer is working who is programmer talking to programmer is going to talk to code reviewer so the code Reviewer is going to create come up with these test cases and review this thing and the code Reviewer is going to not necessarily disc cases the review cases the code Reviewer is checking the code that has been written by the software engineer and it just goes back and forth back and forth it's very very interesting to see because that's obviously what happens in a real software environment I mean people don't work in silos people work as teams and then they talk to each other they give comments to each other and as you can see even in this live replay you would see these agents different roles different roles are there like for example here it is programmer unquote reviewer they would exchange information between them at every single stage and it's quite fascinating to see I think like this has been what I've been thinking for quite a while that you don't or you cannot just build one agent and expect it to do everything you need like a family of agents that is exactly what chat Dev is like it's a bunch of Agents working together talking to each other trying to solve a problem commenting feedbacking and asking them to fix something and that is something that you would see more evidently as we go through this or also when you start trying different projects you would notice this as you can see on the screen there was a feedback to create the requirements for txt not the requirements.txt has been created now there is a new role called counselor and I'm not very sure like what the counselor does it but now once again it has come back to see CEO for reflection on the CEO is going through this and understanding and saying okay let's let's add the requirements.txt which also gets added to the documentation like the final manual user manual that somebody who doesn't even know anything about this project can use it so after all these things you will get a very simple user manual that looks like this explaining you all the details on how you can run or how you can play so for example it says run the main dot Pi file with python main.pi and that game window will open and then you can start doing this thing it doesn't necessarily show you an animation here but the very simple way you can run this thing is you can go ahead and then say enter into the particular folder in the first place once you once you enter into it select the right folder just go inside the warehouse select the right folder and once you go there you can start running the file so all you have to do is python main dot pi and once you do main dot Pi it is going to create the hangman in this case once again there is no animation but it still works I'm going to guess the word and it is quite honestly like disappointing for me like how much time it took for me to guess this word and this is a meta word at this point if you are watching this video I would strongly encourage you to guess this and comment in this if you have managed to guess it like before I run out of all the alphabets on my keyboard if you have guessed it commented I would definitely want to see how many people managed to guess this and I felt actually dumb while recording this video that how stupid I am that I did not guess the word which is quite meta about what we are doing in this place I'm going to just pause this video guess it comment it let me know what is it you're not as lame as me you would have guessed it it's game g-a-m-e the only thing is they didn't close the game properly it didn't give me a score but again I asked for a simple application it's a very simple GUI graphical user interface for me to play Hangman quite amazing quite impressive like I said this is a first agent that just simply ran like no strings attached I didn't run out of tokens even when I had like only 3.5 I didn't run out of uh like the chat limit that you usually get with 3.5 it's quite amazing you can go ahead check the project you can open the code and even the code is like quite really good like you cannot I honestly I cannot believe it's Ai and you can go ahead and then visualize the chain like you can see what all things happened what the face what are the different discussions that happened and it is very interesting to see how this kind of like went through and all these things using GPD 3.5 turbo and the quality of code everything makes me believe that this is quite amazing the fact that you can run a software development team or you can run software development company just using AI is unbelievable it's one open AI 3.5 turbo and it plays the role of CEO CTO product officer coder code reviewer like programmer software tester everything all together it has been like it has the example show it can build a lot more than what we have done it in this exercise but I wanted to make a clear step by step tutorial for you to run this on your own recapping everything clone the repository enter into the repo chat there create a virtual environment activate the virtual environment make sure you have entered the open AI key as an environmental variable once you have done this thing all you have to do this run the file run the Run file if you want to enable the live replay then go to online log and enable it but not then run the file with the task name once you have run the file with the task name then you can start seeing the magic then enter inside the warehouse go into set the project run the file whatever the manual has given I hope this was helpful to you in running chat Dev a true software company just purely run by AI agents if you have any questions let me know in the comment section otherwise kudos to the Developers for making an amazing project see you in another video Happy prompting