Check out Pieces for Developers, which is free for individuals and does much more than GitHub copilot and JetBrains with fewer errors. If you test it, I think you will see the difference.
copy paste from gpt4 & claude 3. i became pretty much a prompt engineer. it's a huge speed boost because my prompts are a) a lot shorter than the code and b) i can skip reading docs, i can directly say "do x in framework y" or "do x but like it's done in ". in my workflow, i don't need fancy autocomplete, i need "given x,y and z, do abc". autocomplete can't migrate or refactor code. autocomplete is not a rubber duck.
As a multi-decade developer myself, I can honestly say this is one of the best articulations of the power and limitations [for now and the near future] of AI coding assistants I've seen so far. Bravo!
Great great mindset. I had the same feeling. AFter the first 'Wooow', i slowly started to realize these tools are by far not perfect. And will create more errors if you dont know what you are doing (as a junior) But as you said, the 'simplify code' is a great way of learning or realizing new ways of coding. Ai coding, for now, is a supplement. Not a complete meal.
You're absolutely right. I comment on my case as a reference. The 'AI Assistant' TRIAL autocomplete was suggested as I typed, so all good. When I hired him, he didn't, and just like Marco says, you have to give him very, very specific instruction, and then he writes a lot more than you asked for, so you have to be reviewing a lot.
I should saw this video 2 hours ago, I already own the Copilot and is very very fast / responsive over the Jetbrains AI, the advantage that I see in Jetbrains AI is IDE integration (commit message, test generation, file creation) but at the of the day as you said, I want the AI to predict my thoughts, not prompt it all the time.
After using copilot for almost a year now and trying out ai assistant for the past week. I completely agree with Marco. Copilot's inline code completion is miles ahead of ai assistant. It even seems to read my mind at times. Unfortunately, AI assistant's inline code completion is barely noticeable. So much that I stumbled across this video trying to figure out if something isn't setup correctly. It's unfortunate too because I prefer ai assistant for all the other features but inline code completely is arguably the most important thing when trying to be fast and efficient.
Since I use IntelliJ, I'd really wish the AI Assistant inline code completion to improve since it integrates more neatly with the IDE. I'm planning to test it every quarter to see if something improves. However, based on what you mentioned it seems it is still the same.
Wise words and well put. The experience can range from really pleasant and a huge boost when a model is very good at a language (like in Python) to a complete drag when generating code by prompting on what a model is not very competent, like what I'm evaluating at the moment for Zephyr RTOS. It's C, but all models I've found are trained on old codebases and a lot changed since it's so the generations don't use the right approach and don't compile without spending a lot of time fixing them.
Thank you 🙏 You reminded me something I didn't mention in the video that puzzles me. Many believe AI will sooner or later perform most of the coding, while engineers will focus on problem solving. However, who is going to create up to date codebases that use the latest patterns, frameworks, and languages which are necessary to train properly these LLMs? I can imagine many scenarios. However, I still think real coding skills will continue to pay off, possibly even more than today if they become rarer to find.
Hi there professor! Wow what a pro presentation here! It seems to me them have somehow the same limitation for everything AI stuff today, and it's reassuring in a way. Is this the dawn of a forthcoming future of dreadful apps/software?
Something I was discussing today with my colleagues at work is that Test Driven Development is even more important now. If you autogenerate code at the very least you would know it's satisfying a test case you crafted. I'm actually surprised there's no such functionality yet and it's inversed. The assistant can generate the test from the code. I think that's the wrong premise. The test should define the result we expect so we should make sure to have at least that one vetted closely before we start trusting what the AI assistant creates. By the way, long time no see sir! Hope all is well! Un abbraccio!
I wish the assistant would read my mind and shut up when I don't want my flow interrupted. I am an advanced developer and it bothers me to know end if something starts typing inline.
I know what you mean. There are times it is very disruptive! Not so long ago I had posted the very same comment. I wish the suggestions were offered only when pressing some shortcut. As far as I know it is possible with Copilot on VSCode but it is not very easy to set it up (at least that was the case last time I tried).
Yeah there are some extra security features if you use the version for business. However it is a delicate topic and I can see there's concern about involuntarily exposing private codebases by using these tools.
It's a very good question which unfortunately it's difficult to answer because heavily dependent on the context where the tech lead operates. I think culture and the size of the team are two important factors. If we talk about a tech leads operating in a single development team, my opinion is that they should code. I would prefer that they lead by example by writing code, defining the architecture of the application and making sure it is respected (mostly through automated tests, e.g. ArchUnit). If we are talking about a team with mostly junior developers, I still wouldn't want them to work as gatekeepers by approving merge/pull requests. I'd rather suggest they favor collaboration with peer / mob programming techniques, i.e. they write code with other team members rather than acting as a supervisor or safety net. I would also put more emphasis on automated testing. The confidence of merging code to trunk or develop branches should come from testing and not code reviews. For what concerns a tech lead that operates across multiple teams, I'd expect his main concern would be defining macro-architectural rules and ensure the overall sanity of the larger system. That would be still more pro-active that review code after it is written. Usually, that's too late and any request for rewrite will cause friction. Hope this helps!
Hi Marco Lenzo , you are doing very well with your video . No matter how great your video is, you will get NO views if you have not done the proper SEO & keyword research for your title, description, tags, and more. do you wanna know more about this ?
Bro, stop rolling your 'r' in English, this makes it EXTREMELY difficult to listen to what you're saying. It's a simple thing you can do to improve your English by literally 50%. If you do this it will really help your channel trust me 👍
Believe me I had no idea that in English the R is not rolled 😅. I just looked online and effectively it's as you said. Thanks for the tip. I'll try to practice it.
@MarcoLenzo Marco, parlerò solo per me, ma il tuo inglese è eccellente: ricco, professionale e molto chiaro. Il tuo accento italiano è un piacere extra per me. Non cambiare nulla!
