Is Github Copilot worth it?

[-]

MrUnoDosTres@reddit

I wonder if Claude or Codex is simply just better nowadays?

[-]

100% yes!! Github copilot is no longer an autocomplete feature. In fact I would argue they should disable it because of the vast amount of agentic integration it now has with the variety of models you have access too. Not to mention rhe pricing much more clear and manageable and really is cheaper then alternatives. Overall github is THE agentic platform for the SDLC.... But... now it is expanding with the github copilot sdk to where now its simply a powerful agentic platform that you can lean on for your agentic apps with access to all the advantages listed above but now in products instead of just the SDLC.

[-]

kobbled@reddit

Super useful in a company with a huge codebase that has many shared patterns to train it on, not as much so for your hobby project.

[-]

booway-war@reddit

Explica mais sobre isso, o Copilot do GIT é bom mas para Projetos grande, eu tenho impressao que por ser da Github ele teria uma documentacao de codigo enorme como base de conhecimento, sera que isso pontua usar o Copilot ?

[-]

kobbled@reddit

I would still try using it! It still may be helpful for you

[-]

LargeHandsBigGloves@reddit

I love using GitHub copilot at work. I'm primarily a SQL and C# developer and I'm using AI to help generate code snippets, ask it opinionated questions, and for suggestions on how to do/implement x idea. I do not use agent, ask it to do complex tasks, or anything like that but it has been a huge time saver - especially when doing things like translating large data formats from one to another as opposed to a manual mapping process. I can skip straight to review on that.
If you're using your AI code to generate everything you do and you want your whole codebase as context, then no, it won't work for you.

[-]

godndiogoat@reddit

Copilot pays for itself when you treat it as a pair-programmer for the boring bits. I keep it on for repetitive data-layer chores: writing EF Core mapping classes, converting messy CSV to insert statements, and sketching out NUnit tests. A trick that saves me review time is feeding it the contract spec in a comment block first; Copilot spits out the method shell with param names that match the doc, so diff noise stays low. For translations I let it build the initial C# switch then immediately ask for a SQL version-cuts copy-paste errors in half. If you hit the context limit, split the file and prompt on just the model section before the logic.

I’ve messed with Postman for quick API calls and Prisma for type-safe DB access, but DreamFactory sneaks in when I need a throw-away REST layer on top of a legacy table.

Copilot is worth it if it keeps you out of grunt work every single day.

[-]

ResponsibilityIll483@reddit

I paid for a year and I barely use it. LSP completions pop-up faster and I end up using those 99% of the time.

[-]

PragmaticBoredom@reddit

Why is everyone talking about paying for a year? They have a monthly plan with a free 1 month trial.

Sign up for the monthly plan. Set a recurring calendar alert to cancel before the renewal date if you haven’t used it enough.

[-]

ResponsibilityIll483@reddit

My company gives us a 50% kickback if we pay for a year

[-]

p_bzn@reddit

Past month I was experimenting with GitHub Copilot and my conclusion is no.

Features I was using: * Autocomplete (completes code as you type) * Chat

Autocomplete is a disaster really. 90% of the time it’s a miss. All what it brought is distraction to the process.

Chat is OK, but somehow just talking with ChatGPT is much faster and I tend to get better results out of it. What I liked is ability to switch between different models while using the same UI.

Still looking for “perfect” setup for myself. Right now the best autocompletion I found is supermaven. I pair it with ChatGPT official UI with projects so context is there most of the time.

[-]

appoloman@reddit

Pretty much my conclusion too, maybe it's just comfortableness, but the GUI for ChatGPT is pretty pleasant compared to integrated solutions.

[-]

LongUsername@reddit

I've had better luck with code generated by Claude in Copilot than ChatGPT

[-]

PragmaticBoredom@reddit

I’ll go against the grain: I barely use my Copilot but the $100/year is trivial for the value I get out of it.

My most frequent use cases: - Writing one test exactly how I want it and asking Copilot to repeat it for a list of different scenarios. - Refactoring one function exactly like I want and then point Copilot at doing the rest the same way - Summarizing code paths through a new database for me while I walk away and get a snack. Saves a lot of grepping and searching. - Enhanced autocomplete. Even if it only works 25% of the time it’s saving me a lot of typing in verbose languages. - Asking to simplify code blocks. Some times it finds library functions that I forgot about or overlooked.

$100/year is rather trivial relative to a $5K MacBook Pro and the thousands I spend on other SaaS and software licenses.

[-]

lorryslorrys@reddit

What is co-pilot better at than an IDE in a browser and does it matter?

1) It knows about the context of your code - Is it useful? No. It doesn't understand much. It can't understand anything beyond snippets in a file, which can also be easily copied.
2) It's in the IDE - Yeah, this is convenient, but not a killer feature.
3) Autocomplete - No. Autocomplete is a trash fire and I keep it turned off. I come back from time to time to see if this feature has risen to the level of being net-positive in usefulness, but again, it's not going to be a killer feature anytime soon.
4) Unlike free AIs, it doesn't take your confidential data for training - Yes. This actually is a killer feature for me, but it wouldn't be if I was just coding my own stuff.

So I would say no. I don't think it's worth as an individual to pay for co-pilot.

[-]

sagiadinos@reddit

Nope, waste of money. I used Copilot pro for about three months. Code completion is ok, but refactoring and code suggestions are mediocre.

And do not dare to write any unit tests for existing classes with Copilot. Ok, except you want to test setters and getters only.

It fails even at the simplest classes: Start mocking private methods or public inherited methods. Or mock even the testing class.

Also nice was unwanted testing of private classes, on the top with deprecated Reflections methods (PHP).

And not forget the inconsistency naming of class variables. In one class it was mockClassname then classnameMock or only Mock or only classname.

Sometimes there where test attributes or setUp methods, sometimes not.

Penetrant ignoring prompts.

I replaced it with Jetbrains Ai using Claude Sonnet and Google Flash which works significantly better (not perfect) for my cases. Around 30 % succeed rate in tests.

Greetings Niko

[-]

wiseaus_stunt_double@reddit

Like others, I have it through work, and it's hit-or-miss, and it way too often will hallucinate the wrong solution regardless of model. Agent mode is a good step in the right direction, but Cursor's MCP agent is better -- even with the same underlying model.

[-]

shugadibang@reddit

The new Agent / edit modes have been nice for delegating small specific changes to it. Its ability to have project wide context has been a grade upgrade from the current “focused file” context.

Also having access to different AI models is nice when I am not satisfied with the default model I use.

YMMV

[-]

Weekly_Potato8103@reddit

The autocomplete is too invasive using the IDEs, but the chat works quite well. Using it with zed editor is really something, especially using libraries or tools I'm not familiarised with.

I don't know Cody but I definitely recommend Copilot. Or better said, any of these new AI tools out there

[-]

thephotoman@reddit

No.

Autocomplete is too aggressive and therefore usually wrong. And the chat experience is about as effective as Googling with site:stackoverflow.com.

As such, there’s one anti-prodictivity element (autocomplete) and one thing that is not a real improvement on its non-AI competition. Anyone who is more productive with AI has been failing to use their existing automation tools.

[-]