text
stringlengths
0
2.35k
**Natalie Pistunovich:** Kind of the other way around from Copilot. I saw in some -- for example in Replit, which is kind of an online IDE, and there you can highlight a function and then say what does it do. And then it will -- similarly, I guess, to a readme, it will be sort of your English used in programming langua...
**Alexey Palazhchenko:** Well, I can imagine it doing pull request reviews. They already have quite some help with linters, and CI, and all that... But imagine Angry AI saying "You should DRY this code" or "You should use single explicability" or something like that... And then there would be a slider, like "Angry team...
**Natalie Pistunovich:** Or like a linter for the answers... A linter before you basically submit your review, that kind of helps you choose the -- "This word is nicer than that word." Or more encouraging.
**Alexey Palazhchenko:** Actually, one I think I would want from Copilot is to generate proper Go configuration file for me... Because it has so many configuration options.
**Natalie Pistunovich:** Yeah. It can be sort of like a smart configuration values, right? Kind of like "This is my use case. Give me what's a good default to have."
**Alexey Palazhchenko:** Yeah. Or just "This is my code. My configuration passes my code." This is how I disabled everything.
**Natalie Pistunovich:** I hope somebody is listening and writing down and making a startup of each of those ideas, because it's all useful tools.
**Alexey Palazhchenko:** Yeah, yeah.
**Natalie Pistunovich:** I have one more to add to our list... Something that you give it a piece of code and then say "Find the security vulnerabilities." Something to point out to you "This is what you should be doing better."
**Alexey Palazhchenko:** Yeah, yeah.
**Natalie Pistunovich:** Would you say that one of the interesting things about FerretDB is that it's actually AI-generated, augmented? The first AI-augmented database?
**Alexey Palazhchenko:** Yeah, you can say so... Actually, when I started I was just alone, myself and Copilot. So you can see he's my co-author now. Right now we have a lot of people from community just jumping in and starting to help... But the core of it was written by me and Copilot. That would be an interesting ta...
**Natalie Pistunovich:** \[28:16\] By machines, for machines.
**Alexey Palazhchenko:** Yeah. \[laughter\]
**Natalie Pistunovich:** I guess it can be also interesting if it helps you to choose names for your project... You know, names for variables. This is a great tool to have. Something that will help you choose variable names.
**Alexey Palazhchenko:** Yeah, that would be very cool. And yeah, names for projects are very hard, actually. We spend a lot of time thinking about it.
**Natalie Pistunovich:** And this can also be aware of all the licenses out there.
**Alexey Palazhchenko:** Yeah. You know you can license particular strings, and then you can use the strings in your code.
**Natalie Pistunovich:** Or an AI that helps you choose a license for your project.
**Alexey Palazhchenko:** I see, yeah. That would be interesting, too.
**Natalie Pistunovich:** The list is really long at this point. I start to forget -- I'm gonna listen to those afterwards and write them down, one by one.
**Alexey Palazhchenko:** The next step would be it would just write a special license for your code. Only for your code. Custom-tailored for you.
**Natalie Pistunovich:** Mind=blown. \[laughs\]
**Alexey Palazhchenko:** It's like, we don't have enough licenses already...
**Natalie Pistunovich:** I guess you can always make it more specific... So what would be your recommendations for Go developers who are not yet using Copilot, or kind of just played around with it a little bit? So how can you use that as a tool to make your developer life better? What can you already do?
**Alexey Palazhchenko:** Yeah, so I would say the hardest problem for you could be just get access to it... Because as far as I know, it's still not public for everyone. If you have it and if you've found a way to have it -- maybe people can contact you and you ask them nicely to give some access... \[laughs\] But I wo...
**Natalie Pistunovich:** So starting by better documentation.
**Alexey Palazhchenko:** Yeah. And maybe you can just start a new project completely and open a readme file and start typing, and write a sentence, and just autocomplete.
**Natalie Pistunovich:** Yeah. Yeah. It can be also interesting to use for maybe refactoring.
**Alexey Palazhchenko:** Refactoring would be actually a really interesting use case... Because right now, all refactoring I know in various editors and IDEs is just completely algorithmic. It's very deterministic. Simple methods like a name or extract might be something like that. But if you're using AI - yeah, that w...
**Natalie Pistunovich:** I can imagine also improving your testing. All the tests that you have for your code; it can also be a quick, easy win.
**Alexey Palazhchenko:** Yeah... For that, I would still prefer fuzzing. But as for generating initial seed data for fuzzing yeah, that would be great. Actually, I think I did write some tests using Copilot this way.
**Natalie Pistunovich:** I'm curious now if there's already some existing library that somebody created that "Here's a great dataset (or pack), all the things you can ever test, all the inputs that are worth being tested", or something like this, and generated by Copilot, and then this will become kind of one of the ne...
**Alexey Palazhchenko:** Yeah, actually there's a repository, something about like "repository of bad strings" which contains all these weird edge cases, like unicode, zero strings, null strings, and all that
**Natalie Pistunovich:** It would be interesting to tell "fuzz it for me", or something like this. Find fuzzy inputs.
**Alexey Palazhchenko:** \[32:02\] That might be another most effective way to use a machine learning framework, but...
**Natalie Pistunovich:** I probably saw some interesting examples, but yeah... Maybe not. I'm trying to think what other interesting recommendations we can give to listeners who are just getting -- or you know what? Let's take one step back and let's speak to those who are maybe a little bit cynical about including AI ...
**Alexey Palazhchenko:** I would say a simple answer is it just really helps. There's some issues with representation, like I said before... For example for me, sometimes these suggestions are not formatted very nicely; sometimes they jump when I start typing and want to take a different direction it suddenly appears. ...
So I would say representation could be better, but when it works nicely, you go much faster. Significantly faster. And that helps. You can be more productive. Not productive in a way as a joke; very productive as in you type a lot and then delete a lot, and you're just typing very fast... But in a good way.
**Natalie Pistunovich:** Well, to play a little bit the devil's advocate, we started this show with listing all the different things that you might need to be reviewing if you are getting the help of an AI to be your co-programmer, if you're doing pair programming with it... So is it actually more efficient?
**Alexey Palazhchenko:** I would say yes. One problem that we did not mention yet - for me at least - is that you just compile it, you have to switch modes between reading and writing. Sometimes you read the code, you understand where you should place new code, how it works, how you refactor it, and then you start writ...
I would say that I did some measurements not scientific totally from me of course, but I've found it helpful... Simple code, repetitive code is generated basically by GitHub Copilot automatically.
**Natalie Pistunovich:** Would you say that Copilot is a good tool to use if you are diving into a completely new codebase? ...so you're kind of even trying to understand what goes where... Speaking of reading versus writing.
**Alexey Palazhchenko:** I don't know... Right now, I don't think Copilot provides much help there. I mean, it works for small pieces, for like "Explain this function." I don't think it works for "Explain this project" or "Explain this module/package."
**Natalie Pistunovich:** Library...
**Alexey Palazhchenko:** Yeah. Who knows, maybe that will work in just a few weeks, months, maybe a year.
**Natalie Pistunovich:** Yeah.
**Alexey Palazhchenko:** I think one Go-specific way where Copilot would help is as a generics replacement. You don't have to write all this code yourself, you don't have to use generics; you can just ask the compiler to generate all the various functions for different and it works great.
**Natalie Pistunovich:** \[laughs\] That's an interesting approach.
**Alexey Palazhchenko:** Yeah... I mean, we used these code generators everyone wrote back in the day, and... reflection, interface...Just use Copilot. That would be fine.