Yay @AaronAsherRandall! I've certainly developed a lot more Java apps than any other language in my career. Java's nowhere near "dead", as many have proclaimed for about a decade now. Tons of enterprises still roll out brand new platforms and services based on it. Hope your REST API was fun to build with AI!
Great video! I have just been using Aider for the last three weeks to create an entire web application and I am not a developer. I use aider to first create a Playwright automated UI test of what I want to achieve and then ask it to implement the code, for making the test pass. It is working pretty well. Then you are able to catch regression test errors. I implemented approx. 50 users stories, which took around 1-3 hours a piece. But a few of them blew everything up and took maybe 10-20 hours a piece. I have been thinking about making the frontend first, mocking the backend. And then when that is working, get it to create the API and backend. What are your thoughts on that?
Thank you so much @KennethKammersgaard! You sure you're not a developer? You talk like one 😉 Sounds to me like you're rockin' it! And, regarding your idea... hadn't thought about doing that in reverse. So, would you first use AI to create a "mock" backend, then use AI to generate a frontend against the mock API? If so, sounds like the only difference might be that you're trying to use the frontend to flush out your backend design and requirements? Versus the other way around as I did - using the backend to drive the frontend features? No reason at all I can see that won't work. The main difference here is a matter of how far you go with developing the full backend early on. Keeping the backend agile and light - i.e. by not yet implementing full functionality, just mocking it - would help ensure you don't waste a lot of time on the backend up front, only to probably make major and deep changes to it, as you learn more about the real needs of your frontend UI. Am I getting you right here? If so, I'd recommend still doing something similar to what I did in this tutorial - i.e. scaffolding the REST API first. BUT, only have aider implement the "REST controller" layer AND just tell it to mock up those REST endpoints. So, they all return an appropriate, but canned response. Then, have aider generate the OpenAPI docs, as I did in the tutorial, so you can feed those to aider (using aider's /web command) to guide aider in generating the UI's API integration. That approach saves you from implementing all the real API code - like the Service layer, the repo, etc. If you discover new reqs or tweaks needed to the API as you build out the UI, it's easy to go back and have aider make updates (still using mocked endpoints) and update the OpenAI docs, etc. I hope that all made sense?
Thank you for that @jlyunior! I haven't gotten around to Claude Dev yet. I will but can't say when. I hear great things about it. If you give a whirl, I'd love to hear your thoughts.
Hi @savastutumlu1566! When I'm generating a brand new app like I do in these tutorials, hands-down, aider's my go to. Also, when I'm making updates that's likely to require changes to multiple files at once - aider. When I'm doing things like "add doc strings or JavaDocs to this file" or "explain this code to me" or I'm in a source file and working on it directly - say, I'm in a flow and editing the file to add a new function or refactor a function... those are the scenarios where I usually flip over the Continue. Does that make sense?
I really appreciate your feedback @bgmspot7242! I do want to get to this topic. That will be a LOT to try to cover in one video. I am planning on doing one soon comparing and contrasting the two very different categories of AI coding assistants - i.e. the terminal-based, IDE-agnostic approach aider uses vs the IDE-integrated approach most other assistants use. There are so many coding assistants now that it's a really tall order to do justice to a comparison in one video. I'll put some thought into how best to attack this problem.
Hey @comosaycomosah! There are likely some coding assistant options out there I'm not yet aware of. But, of all the ones I've used - aider, Continue, Cody AI, Codeium, Cursor, etc. - aider's probably your best bet for any coding environment that's not either VS Code or IntelliJ. Because it's terminal based and can work directly on files and directories (and manage git commits for you), IMHO aider is by FAR the best coding assistant when you don't want to tie yourself to an IDE (or even when you do, in many cases). Free however... because nothing yet touches Claude 3.5 and it's well beyond any other LLMs for coding right now, PLUS the fact that you're working outside of VS Code, I'm not aware of any free option that I'd recommend. Having said that, you could try Llama 3.1 on Groq with aider. With the free tier, I'm guessing you'll run into unacceptable rate limiting that will drive you crazy if you're coding much.
Binge watching your vids! Great content 🎉
Thank you so much for your kind words @MagagnaJayzxui!
Lovely stuff brotherman ! You deserve blessings for this
Thank you so much @jefferystartm9442!
Wow. Awesome workflow! And thanks for shining light on Java! This inspired me to make a rest API as well in Java!
Yay @AaronAsherRandall! I've certainly developed a lot more Java apps than any other language in my career. Java's nowhere near "dead", as many have proclaimed for about a decade now. Tons of enterprises still roll out brand new platforms and services based on it. Hope your REST API was fun to build with AI!
Brilliant. Your videos are thoughtful and express your skills. Thanks for sharing.
Thank you so much for your kind words @alane1988!
Not a coder but this is super interesting to actually see the process
Thank you for that @vitalis! A good time to become a coder 😉
Great video! I have just been using Aider for the last three weeks to create an entire web application and I am not a developer. I use aider to first create a Playwright automated UI test of what I want to achieve and then ask it to implement the code, for making the test pass. It is working pretty well. Then you are able to catch regression test errors. I implemented approx. 50 users stories, which took around 1-3 hours a piece. But a few of them blew everything up and took maybe 10-20 hours a piece. I have been thinking about making the frontend first, mocking the backend. And then when that is working, get it to create the API and backend. What are your thoughts on that?
Thank you so much @KennethKammersgaard! You sure you're not a developer? You talk like one 😉 Sounds to me like you're rockin' it! And, regarding your idea... hadn't thought about doing that in reverse. So, would you first use AI to create a "mock" backend, then use AI to generate a frontend against the mock API? If so, sounds like the only difference might be that you're trying to use the frontend to flush out your backend design and requirements? Versus the other way around as I did - using the backend to drive the frontend features? No reason at all I can see that won't work.
The main difference here is a matter of how far you go with developing the full backend early on. Keeping the backend agile and light - i.e. by not yet implementing full functionality, just mocking it - would help ensure you don't waste a lot of time on the backend up front, only to probably make major and deep changes to it, as you learn more about the real needs of your frontend UI.
Am I getting you right here?
If so, I'd recommend still doing something similar to what I did in this tutorial - i.e. scaffolding the REST API first. BUT, only have aider implement the "REST controller" layer AND just tell it to mock up those REST endpoints. So, they all return an appropriate, but canned response. Then, have aider generate the OpenAPI docs, as I did in the tutorial, so you can feed those to aider (using aider's /web command) to guide aider in generating the UI's API integration.
That approach saves you from implementing all the real API code - like the Service layer, the repo, etc. If you discover new reqs or tweaks needed to the API as you build out the UI, it's easy to go back and have aider make updates (still using mocked endpoints) and update the OpenAI docs, etc.
I hope that all made sense?
What about Claude Dev? Which one is better? Can u make a review or a comparative ?
Thanks for your videos and content ! Keep going !
Thank you for that @jlyunior! I haven't gotten around to Claude Dev yet. I will but can't say when. I hear great things about it. If you give a whirl, I'd love to hear your thoughts.
Good content.
Thank you for that @onyangomwangi3495!
Patiently waiting for more. Really impressed by your workflow.
Would you say continue or aider is your preferred choice for generating apps ?
Hi @savastutumlu1566! When I'm generating a brand new app like I do in these tutorials, hands-down, aider's my go to. Also, when I'm making updates that's likely to require changes to multiple files at once - aider. When I'm doing things like "add doc strings or JavaDocs to this file" or "explain this code to me" or I'm in a source file and working on it directly - say, I'm in a flow and editing the file to add a new function or refactor a function... those are the scenarios where I usually flip over the Continue. Does that make sense?
this feels like an 8th grade project, have you gotten any success on enterprise level stuff where you use custom frameworks not springboot?
Please do a video on cursor vs aider vs claude dev vs continue
I really appreciate your feedback @bgmspot7242! I do want to get to this topic. That will be a LOT to try to cover in one video. I am planning on doing one soon comparing and contrasting the two very different categories of AI coding assistants - i.e. the terminal-based, IDE-agnostic approach aider uses vs the IDE-integrated approach most other assistants use.
There are so many coding assistants now that it's a really tall order to do justice to a comparison in one video. I'll put some thought into how best to attack this problem.
How much did it cost? I know aider has some promt optimizations, still?
Hi @justafreak15able! Please see the video I published with the title: "AI Coding Assistant Costs Revealed".
Is this prolly the best free option? Im kinda looking for something that works with nvim
Hey @comosaycomosah! There are likely some coding assistant options out there I'm not yet aware of. But, of all the ones I've used - aider, Continue, Cody AI, Codeium, Cursor, etc. - aider's probably your best bet for any coding environment that's not either VS Code or IntelliJ. Because it's terminal based and can work directly on files and directories (and manage git commits for you), IMHO aider is by FAR the best coding assistant when you don't want to tie yourself to an IDE (or even when you do, in many cases).
Free however... because nothing yet touches Claude 3.5 and it's well beyond any other LLMs for coding right now, PLUS the fact that you're working outside of VS Code, I'm not aware of any free option that I'd recommend.
Having said that, you could try Llama 3.1 on Groq with aider. With the free tier, I'm guessing you'll run into unacceptable rate limiting that will drive you crazy if you're coding much.
Title should contain “without YOU writing any code”
Thank you for that @thyagtubes! Yes, technically. Unfortunately, the title's already a little too long as it is 😆 It gets chopped off on some devices.
Use cursor and it’s pretty amazing
Zed ai also good
@@xelerator2398 I just saw the post on Zed and AI. That's going to be interesting. Cursor vs Zed...
Hey @lloydjose8137! Yep, lots of great buzz around Cursor. Seems to be a really solid option.