I Tried Every AI Coding Assistant

Key Takeaways

In this video, we're going to go through five different AI coding assistants to find out which one is best for you. The first one that we're going to go through is GitHub Copilot. Now, personally, this is the first one that I started with and just to test it out baseline, I'm going to ask it to generate a function. Palm basically means the string is the same forwards and backwards. Okay, so it created it and I'll say okay, insert into the test editor. That was actually pretty quick. code readability. I mean, if you know Python, this is pretty standard. First, it converted it to lowercase, then replaced any open spaces with empty string and basically checked if we reverse it by doing slicing operator and to see if it's true. Now, this actually pretty smart because personally, I didn't think at first to convert like lowercase into account for spaces. So, that is pretty good. Now, I do wish that it had already added test cases to actually see what the true, false, and edge cases were going to be like, but I mean, this is pretty good for Python. It didn't need to specify the type of like the string or the boolean. So, GitHub Copilot, good job. I'll say generate test cases. So, all these test cases were added and it even created the infrastructure to run the test cases like through this like for looping. It has a bunch of different ones. Some just small strings, some sentences, empty string. It accounted for that case as well. So, this is actually pretty good. One cool thing about GitHub compiler is you can add comments. Account for null strings or random inputs. Let's see what it does. Oh, sweet. Cool. This is what I really like about GitHub Copilot. You can just put in the comments and then it'll kind of like fill out the function for you. So this is good. We have a little chat editor that we can talk to in English and then we put in a comment, it can fill it out. So code readability is really really strong here. Now I'll say this execute the test cases. Now one thing I didn't like is it didn't automatically update the test cases to test out for the null scenario, which kind of sucks. Okay, so all the tests passed. Good job. I'll say account for null scenario in test cases. All right. Well, it added a line. So, I do got to give props that it was able to account for it. But ideally, the AI should already be able to do that. Like, I shouldn't have to tell it additionally. Now, something important to note is like Copilot, it's running on the GPT4.1 model and you can like switch out the model. So, maybe different ones can improve the performance. Honestly, I should probably be using GPT5 since it's out, but we'll save that for the later in the video. The next thing I want to test it on is a standard elite code problem. And so, what I'm actually going to do is I'm going to take a screenshot of this problem. For those who don't know, this twosome problem is basically like you're given an array. Can any two numbers from this array add up to a target sum? Okay, I'm just going to add the screenshot. Solve the twosome problem and remove the palendrome code. Let's see how quickly it can do it. Oh okay. Let's see what it did. Wow. So it used a dictionary. It enumerated nums by basically having an indicy in the value. Goes through it calculates the complement which is actually how you keep it O of N because otherwise like the brute force approach is O of N squared where you do a nested loop. What is the time and space complexity of this? and run the code to see if it works. Yep, it it matched the outputs to what was expected. Time complexity O of N. Space complexity is O of N because we used the hashmap. Interesting. Is there a way to make it more space efficient? Okay. Yeah. So the way that you make it more space efficient is by doing the brute force approach. However, that scales the time complexity up to O N^ squ, which we actually don't want to do. So, this is pretty good for understanding and also beginner friendliness. I really like it because I mean, if I didn't know and I could just like put in comments if I can just talk to this editor, it is pretty excellent for that. The next thing that I want to do is actually give it incorrect code and see how quickly it can identify bugs because what good is AI if it can't even find out our errors. So I'm going to feed this in this is even function. Oh, okay. It can already identify it's an incorrect implementation. But let's see what is Oh, okay. Quick. Wow, that was really quick. This is really good. So the is even function, I actually had it modulus equal to one. That's actually how you tell if a function is odd. But what it suggested right away was to convert it to zero. And then it started generating test cases. Where did they go? Perfect. This is good. It's still a simple function, so this is kind of baseline expected at this point, but overall, I'm pretty impressed with it. Now, let's try to scale it up even further. I'll say okay, so it's asking if I want it to be local, yes, local, and React. So, it's giving me a list of things to install. Now, I've used other code editors where it just starts automatically going through all that stuff. Run dependency installations. Continue. Continue. Okay. So, there were issues. Would you like me to Yes. I don't like how it's asking me back and forth so many questions. I wish it would just kind of automatically resolve everything. Yes. Retry. So, it's just a bunch of dependency issues now. Yes. launch. Okay, so the attempts failed, but it's suggesting things to try again. Fix the issue for me. All right, so it's telling me to install more things. Yes, it looks like the front end is starting looking. All right, cool. Can't be reached. Come on, Copilot. What are you doing? Saying it's starting up, so maybe I'll give it a second. Reload. Reload. Reload. Reload. Reload. Nothing is working. I can't see it on port 3000. Check if it's working. This is just taking a really long time. And to be honest, it's just not ideal. All right, let's try again. All right, it's still not loading and the commands are taking really long. I don't think Copilot really knows what to do. I don't think it's optimized for these interweed dependency issues. And so overall, when I'm looking to rate these tools in terms of ease of use, I would say it was pretty easy to use. It's integrated into the IDE. I really like the common features. I really like how it made automatic updates. I've heard sometimes some people said it was annoying, but in my history of using them, it's been pretty good. So, I'll give it like a 9 out of 10. Reliability, how often does it give working code on the first try? So, for basic level tasks, it's really good. But when you're giving like advanced level things such as like this project and how it has to deal with many different things, it does struggle quite a bit. So I'll give like a five out of 10 on that. In terms of explanations, it is really good at giving explanations because when it generates code, it give an explanation as to what that code was for like the palendrome, for like the lead code question and even like the time complexities. So I'll give that 8 out of 10. Productivity and speed, that was really fast, but also like really slow for the project. So I'll give like a five out of 10 for that. And then ideal use case, I would say this is best for beginners. when you're doing like basic level tasks, not advanced projects, it is phenomenal. So overall, I would put this maybe like a B tier. Next up, we got the famous Chad JPT. And I'm going to ask it the same exact things. Generate a function to tell if a string is a palendrome. Let's see how quick it can do. Okay. Okay. All right. So, I do got to say it does use slightly more complicated code where it's like this empty string join and it's like splitting it up by characters and then joining them and then cleaning up. And this is al num. I don't even know what it's doing. Maybe it's doing the same exact thing as the other one. I do like how it does generate test cases right away. But I mean, one downside of this is you can't really execute this code. So, you would have to use VS Code or some other IDE to actually execute it. But for what it's worth, this was decent. Maybe a little complicated for beginners, but it does the work. I guess that's what matters at the end of the day. Next up, I got the screenshot of the Tusome lead code problem, and I'll say solve this problem for me. So, it pretty much actually created the same exact implementation as the copilot one. It created a dictionary to store the numbers in the index, and it enumerated it, went through it, O of N, time, and space complexity. Honestly, this one is a little weird because it could just be like taking this answer from like an existing source because this is a pretty popular question and it also does explain right away. I didn't have to ask it the time and space complexity. So, pretty good on that. All right, the next one is the debug checker to see this function. I'll be like, what is wrong? Or actually, I'm not even going to say what's wrong. I'm going to see if it can detect it. Okay, cool. So, it actually quickly identified that the function is wrong. I didn't even have to say debug it. You can just throw in code and it figures it out. It shows the corrected version. It explains why what went wrong and it even gives you test cases to test the true or false of it. But the one bad thing is once again you can't really execute it in here and going back and forth between chat GPT and your IDE, especially when you're dealing with complicated projects can get kind of annoying. Trust me, I've been there. The next thing I'll say is build me a chat GPT clone in Python and JavaScript for front end. All right, let's see how this does. I've never had it actually create a full project because once again, Chad GPT on its own isn't really optimized for things like that, but you don't know unless you try. By the way, this is running on GPT5, so this is like the latest model or at least at the time that this is being recorded. And let's see what it comes up with. Okay, so it gave me a couple of different files and it's telling me how to structure it if I do copy it over to an IDE. So it has the back end uses GPT40 mini asks me to put in my API key base URL index HTML file which is good and it shows you exactly the command you need to run it and it gives you a couple different notes because chatbt is separated from like your IDE so you'd have to go back and forth. I do have to say ease of use is probably like a three out of 10. I do like it. The UI is pretty clean, especially I think after GPT5 came out, I felt it got a little more like smoother, but still for daily coding tasks. It's not optimal. In terms of reliability, it gave good code on the first try. I have no complaints in terms of that. So, I do have to say maybe like an 8 out of 10. In terms of explanations, it did a really, really good job. Before I even asked for explanations, it offered explanations. So this would even be higher than C-pilot for that regard. So probably like 9.7 out of 10. I don't think I'll give it a 10 out of 10 just yet, but we'll see what happens. In terms of productivity and speed, it did this pretty quickly. Even like this whole like app generated pretty quickly and gave me all the steps I needed to run it. I'll probably give it like 7 out of 10 for that. So overall, I think is pretty good, but because it's not integrated into the IDE, I do have to put it at a C tier. I'm sorry. Like the ease of use is just not there. I know programmers are lazy. We don't really like going back and forth. So, I'm looking out for you guys. Next up, we got Replet. And I am really excited for this one. We're going to open this app. We're going to put it in Python. Keep it private. Create app. See what goes on. So, first thing we're going to tell replet is create a function to tell if a string is a palendrome. I totally butchered that. I do got to say for a palendrome check, it is taking a bit of time. Oh, wow. So, it created the code and it ran the test cases and it quickly opened up the console. So, like I basically barely saw the code which is getting really close to like the whole vibe coding idea and it wrote out 49 lines of code. It did this iterated thing like similar to what chat GPT did. I'm not a huge huge fan of it, but it did leave good comments around it and it ran test cases right away. So, I do like that as well. Next thing is solve this to some lead code problem. Yeah. Okay. So, it find out exactly what it needs to do. It's kind of like in an interview where you have to talk through what you're thinking. It does that pretty well. Shows you exactly what files it's editing. It generated the code, generated test cases, executed the test cases. I didn't run a single terminal command, which I frankly hate terminal commands. And so, it did it like that, like that, like that. Plus, it explained how the solution works. mentioned the optimal O of N time complexity time. So that is all phenomenal. Now the true test of the game where the other AIs have struggled so far is creating a clone of chat GPT and let's see what it can do. Create me a clone of chat GPT with a nice UI. All right, so it's creating me a plan, creating a web application, creating a task list, it's using Flask, it's using OpenAI. Okay. Wow. So it created the clone and it did that pretty quickly. This is actually what replet is really good for creating like these quick things that you can test out very quickly. I didn't have to once again deal with any terminal commands, any like local hosting or anything like that and be like what is 2 + 2? Oh, it needs my open AI API key. So, what I'm going to tell it is I don't have an open AI API key. Can you mock the function calls? Oh, it even give me like infrastructure to put it in. If you can handle this and create like a little mock, you get a really huge value ad because you can really spin up quick projects just like that without having to deal with literally any complications. Not many I tools even do that stuff. I put in two messages so far and I have this already. Okay, so it looks like there was some diagnostic issues. It took a screenshot of that whatever like that console issues or UI issues was and it's rerunning it. It's fixing it on its own without me saying anything. And now it's running again. What is 2 + 2? Okay, it basically created a response somewhat. So I can just like kind of test out the UI, see how everything handles. I can clear chat history in case I want to hide anything. I really like Replet because not only can you create these things, you can also like deploy things really quickly to what your specifications are. And I recently created this one website where it's just info about me. And in fact, for like a personal website like this, a while back, I paid someone like about $1,000 for. And I got to create this whole website for literally the $20 monthly subscription on Replet plus whatever like the hosting cost is. And I got it done within just a few hours. And that is the true power of AI. So in terms of ease of use of all this, you don't have to switch through tabs. You can just develop things really quickly just like that. So I would give this like a 9.5 out of 10. You barely have to even talk to it because it can kind of iterate on its own which is a true agentic workflow in AI which I really like. In terms of reliability, it gives you good working code on the first try. Does the test cases executes the test cases and even if it doesn't get it, it retries on its own. So reliability once again 9.5 out of 10. Explanations, it does give good explanations. It explains everything that it's doing. Creates a whole list. It gives you a task list. It gives you a explanation list. 9.5 out of 10. Productivity and speed. Now I do got to say for some of the basic functions it did take a little longer than I expected like the palendrome check to some leak code problem on that I might knock it down to like a seven out of 10. Best fit ideal use case is this is definitely something not necessarily for beginners. Obviously beginners can use it and take advantage of it. But the true value is when you already know how to code and you're just trying to accelerate your workflows. You're trying to launch projects really quickly. If you're a complete beginner, this just hides so many like layers and it's so abstract that I don't think you'll learn very well if you just use this. But overall, I'll put this in the S tier. Next up, we got cursor. And this is something I've recently got more and more familiar with. This itself is an IDE, like cursor. It's like something you download and instead of VS Code, you can like use this to do all your coding tasks. And so it has like this chat feature kind of similar to Copilot. Yeah, let's get started. So create a function to tell if a string is a palindrome. Oh my god, I keep spelling that wrong, but it can correct it. It plans out what it does. Wow, that was fast. Oh, it gave me in JavaScript, Python, and gave examples. Okay, that is pretty good. Insert the Python code on the file. In this case, it actually handled an edge case to check if it's a string in and of itself. This is something the other AI tools didn't really handle super super well. And let's say generate test cases. Okay, so it's going to create a whole new test palm file. 48 lines. Wow. And it's testing a lot of mixed things. And it gives you the command to run it. I'll say run the test. So it failed. Oh, it's even handling other languages. It's handling French. Wow, this is crazy actually. And it's fixing up the test cases accordingly. Okay. Wow, it passed. This is what I really like about AI tools. It shouldn't just do the bare minimum and have it pass or fail. It should as soon as it fails, debug it automatically. That's what true value of AI cuz like if you can just generate code, that's almost like it's textbook teaching you what to do. But the true value for programmers is that iterative process when you go fix it up, figure out what's wrong, that's what really accelerates development. Because a lot of times software engineers, we aren't just writing brand new code. We're fixing existing code. So fact that this can handle it when things fail is a big plus point for it. All right. Now we got the lead code problem. Remove the palendrome code and solve this to some problem. Deleted the files. Generated a new file here. Okay. Once again, it did the thing where it uses a hashmap. It does the most efficient O of N approach. Now, it does raise a value error, which is actually something the other ones didn't really do. Created many different test cases without even me asking. Ran six test cases. Okay, it failed. Let's see what it does. So, in this case, it just got the indices wrong. So, it set up the test cases incorrectly. Perfect. Now, something I like about cursor is you can highlight something and add it to the chat as context and say, explain this line. So I highlighted the value error line and you can just ask it to explain certain specific context and the more specific you can get the better the response is. And so it gives why this value error exists, how it works in practice and examples of when it might get triggered. So now let's do the debugging test to see how it deals with this function. Okay, so as soon as I highlighted it, it gives a suggestion, but I'm going to add to chat fix. Found the issue. Cool. did that pretty quickly. Standard. Yeah. And it gives you the things. It didn't generate full-on test cases. It did for like half second, but then it just Oh, there we go. So, that's pretty good. And it also generated the is odd function. So, it not only generated the test cases to ensure what I wrote was correct, but it also generated what it thought the original implementation was for. So, that is thumbs up. Good job. The last thing we have to do is generate a chat GPT clone. All right. So, it created a to-do list. All right. So made an index file. I think there is a way to automatically not have to say keep all. But now for safety checks, you want to include them. All right. So it tried to install pip and it uh failed. So it's creating a virtual environment. Just kind of like rerunning, rerunning, rerunning. I love not having to do terminal commands on my own. Oh my god. All right. So this is taking way too long. Not sure why it can't handle these terminal commands and servers. And maybe there's something back and forth going on that it doesn't have insight on. Maybe this is something that they should be fixing overall in terms of ease of use because this is a holistic IDE with code and like the the chatbot all like integrated within it. Good work. N out of 10. It's cursor's own like IDE. So this is great reliability. How often does it give working code on the first try? It did this pretty good for this last one. I think the only issue was getting it to run. So like terminal commands, setups, and all that stuff. So reliability, I'll give it like a 7.5 out of 10. explanations. It did a pretty good job of like giving a solid task list when it was like doing a lot of the code and even like when it was explaining the elaborate test cases for like the debugging and all that stuff. So, it did pretty good on that end. I would give that 8.5 out of 10. Productivity and speed, I personally have used this a lot at work. And I know I may not have like demoed it like as like great as functionality is, but I know how like great it is in software engineering environments specifically like when it's like modularizing code. I liked how it made like a separate test case file and that's essential in software engineering workflows for like proper productivity and just clean code organization. So I would say 10 out of 10 best ideal use case I would say like enterprise software engineers those like use cases phenomenal. So I would give it a tier. All right next up we got claude and I already know what you guys are thinking. Oh this is just like ch but just like anthropic instead of open AI. However they are one of the most popular ones. So I think it would be an injustice if I didn't at least use them in this video. So the first thing I'll say is create a function to tell if a string is a palindrome. Yeah, I am not good at spelling. Okay, so it created three functions. One is like the baseline check. The second one is code cleanup. The third one is using two pointers and seeing how it goes after the code is cleaned up. And it runs test cases for all of this. Plus, what's really nice is you can actually publish like these artifacts like how it's on the side tab. It's on an artifact. So, it's almost in a sense like other people can access it if you like generate like a website or some document or things like that which is really cool about claude. I think I've used claude a lot. So, this was a little impressive to see all these things cuz I asked for one function. The next test is delete code. So, remove palendrome code and solve this to some problem. found the optimal solution. It indicated the time and space complexity up here and then it created the brute force and then it created test cases for this as well and it gives you the explanation of the optimal solution, the brute force solution and why the optimal solution is the optimal solution. This is really impressive. Okay, now I'm going to test how well it can debug. So kind of similar to cursor, it indicated right away that it was wrong. It indicated what you should do for the is even and what you should do for the is odd creator test cases. But once again, kind of like chat GPT, you can't really execute anything in their application. Create a clone of chat GPT with a nice UI. Let's see what it does now. Okay, this looks very similar to the rep. Oh, okay. Okay, it actually ran the code here. Hi. What is 2 + 2? Okay. Yeah. So, it definitely did mock the chat GPT thing. The thing is it's good that it did that. However, it should have asked me if I wanted to input like my API key like Replet had asked me. That's one downside. However, it is nice that you can actually visualize it compared to for example like Chad GBT where it didn't even give you that option. Overall, if I had to rate this ease of use, because it's not integrated into the IDE, you would have to switch back and forth, but you get to do quick little baseline testings kind of like what we did here. I would probably give it like a five out of 10 because you can do some here, but you do have to switch back and forth. Reliability, it did give really good code on the first try with a bunch of different test cases. I was really like mindbggled when I saw it created all those like different functions calls without even me asking. 9.4 out of 10 explanations. It gave all the support that I needed to understand what was going on, explain the time complexities, explain why something was better. 9.5 out of 10. Productivity and speed that was really fast. However, because it didn't really ask me what I wanted in this case, I do have to like dock points. It didn't ask me about like the API key thing that I had mentioned and like the fact that you would have switch back forth. Seven out of 10. And then best fit ideal use case. I would say beginner to mediumish developers where you're trying to advance your skills a little more. you're still like understanding a little bit and you want like some quick demonstrations like this and maybe just like full-on code understanding. Maybe you're studying for your coding interviews and you want proper explanation. I would say for that. So, not complete beginners, but also not super super advanced people either. And so, overall, I would put it in the B tier tool to use. Well, that's about all I have in this video. I really hope that you guys enjoyed it. And if you did, make sure to hit the like button, subscribe if you haven't already. If you're interested in my absolutely free tech newsletter, link for that will be in the description below. And if you're interested on is it even worth learning to code in this age of AI, you might like this video right here.