first impressions of the Copilot AI service — Development on

Some developers say that the service makes the routine easier, others say that it often offers a lot of extras.

On June 22, GitHub opened access to Copilot, a “programming partner” for all developers – a subscription costs $10 per month or $100 per year. The service has a trial period of 60 days. The free version will be available to students with the GitHub Student Developer Pack and developers of popular open source projects on GitHub.

What users say about the service.

I tested it for two weeks as I had the opportunity and I can say that it works pretty well for the same type and simple things, especially if you need to write a bunch of unit tests or something trivial.

But as soon as it comes to something difficult and specific, hints of some kind of game begin, since there is no context and requirements. Sometimes it suggests something completely ugly that you have to refactor right away, otherwise you will be covered with affectionate words on the review. Sometimes the proposed solution is the worst possible in terms of optimization. user

Nobody warned me that Github Copilot can auto-complete Markdown documentation, including in Russian.

Well what can I say, this thing already knows everything that a programmer needs to know.

I tried to write on it for about a month. Not bad for beginners, good for coding. Unfortunately, you have to edit and simplify a lot after it. But at the start on the knee, quickly throwing in an approximate direction is a great solution if you understand that 80% will have to be thrown out in the future. In short, to start – a must have tool. user

Unexpectedly, at the end of March, access was opened to me. Tested. In principle, the main conclusion is convenient as adding another monitor when you already have two. Of the super conveniences, it’s good to write unit tests, for me they are boring, but here they throw it right in paragraphs.
And average amenities – good ideas when you don’t know where to start. Uses the code of your project – substitutes more often approximately correct constants and expressions.
Of the inconveniences – sometimes it infuriates you to delete what he writes to you, you delete, and he again enters the same. But apparently a bug.
In general, it is convenient, it helps to get rid of the routine a little, but it will not replace the programmer. user

@varjmes I’m my experience the speed of actually writing code rarely matters and good devs spend more time e.g. collaborating with designers, and a tool like copilot can’t auto-fill domain specific code which is most code people touch in their everyday work

In my experience, coding speed is rarely important. Good developers spend more time collaborating with designers, and a tool like Copilot won’t be able to write the domain-specific code that people most often encounter in their day-to-day work.

I have been using 2 weeks in Visual Studio 2022 with C# language. It helps a lot, often writes correct code. He copes with algorithmic tasks with a bang.

I will probably buy. user

Cool service, sometimes it predicts the future, I have not yet figured out what to write next, but it already offers a function for transforming data from the backend for my front, in 80% of cases it is correct. But I’m afraid to go dumb with him. 🙂 user

I write in GoLang, I cry with happiness! He literally reads my mind on a large code base, I keep typing the beginning of a thought, and he guesses absolutely everything (even with 1 letter): code, comments, name and style of errors. And he doesn’t just guess, but writes in the style of my codebase… I’ve already lost the Tab key!

The only negative is the delay between the input and the appearance of the hint, which, in my opinion, is slow (after all, it has the entire codebase).

“Habr” user

In my experience the biggest help Copilot has done for me was showing me exactly where I was writing a lot of boilerplate code so that I could just simplify it.

It’s pretty cool when it works and you can just tab a bunch of code but most of the time I spent just disabling it.

From experience, the biggest help of Copilot is that it showed me exactly where I wrote boilerplate code, and I could simplify it. Pretty cool when you can just press Tab and get the code, but most of the time I just fixed it.

@frankdilo It’s way too early to charge money. Copilot produces too much garbage. Sometimes it’s obvious garbage but often the bugs are hard to spot.

I don’t get why ppl praise it that much. Sure, in a few years jt might be fantastic but it doesn’t save any time as-is in my experience.

It’s too early to take the money. Copilot creates too much garbage. Sometimes it’s obvious that the code is garbage, but often bugs are hard to find. I don’t understand why people praise him so much. Of course, in a few years it can become cool, but now, in my experience, it does not save time in any way.

On June 3, Flutter developer Evgeny Saturov told in a thread how he and his team tested Copilot for a month.

Reply to post @scoumbourdis

We tested Copilot in one of our projects for a month, eventually deciding to abandon it as a must-have tool. Despite the fact that at the start of the experiment, he struck us to the core.

When the first charm passes, the outlines of harsh reality appear.

The main problem that we have identified is that Copilot is confusing, sometimes offering completely insane solutions.

You can’t help but pay attention to hints while trying to make sense of what is being proposed, although at this time you may have a complete picture of the future code fragment in your head.

Of the uniquely useful areas of application: writing code in an unfamiliar language.

Flutterists often get into the native and write small code snippets in Kotlin and Swift, which they don’t know as well as Dart. Copilot greatly speeds up the process and improves the quality of the code.

It is also noticed that Copilot really learns even from the code base of the project itself. If you write the same boilerplate regularly, it will learn to write it for you.

It is very comfortable.

Copilot is much more useful when writing new code than when making changes to existing one.

The less context around, the higher the chance that the proposed solution will be appropriate.

We assumed that Copilot would greatly speed up writing tests. These expectations were justified.

Good naming is a guarantee of tests that are written by themselves 🙂

What Copilot really teaches well is correct and understandable naming.

Despite a number of really outstanding features of Copilot, the result is sad. It doesn’t speed up development.

After three weeks of experimentation, half of the developers on the team turned off Copilot, continuing to work the old fashioned way.



Leave a Reply

Your email address will not be published.

This site uses Akismet to reduce spam. Learn how your comment data is processed.

On Key

Related Posts