Exactly, my company provides license for copilot and I use it, and while it has some highlights most of the time it actually is more a nuisance than help.
It especially annoys me because it hijacks autocomplete based on types with is own that frequently has subtle bugs, so now if I have it enabled I need to be on guard all the time. With the traditional autocomplete I could just trust it to be correct.
This is my experience. It saves a bit of typing sometimes but that’s probably cancelled out by the time spent correcting it, rewriting nonsense it produced, and reviewing my corworkers PRs that didn’t notice the nonsense.
It will insert comments to explain the more nuanced aspects of my code for those viewing it for the first time
Doesn’t make up for the annoying-ass auto complete hijacking though. Stupid thing keeps making up non-existent functions and api’s and inserting them all over the place.
You have to put it in Ask mode so it doesn’t touch your code also ChatGPT models are free so if you want to ring up an AI bill use the Claude and Sonnet models.
Turn off the autocomplete, it’s shit. Do use agent mode for targeted tasks that are easy but laborious. Don’t give open ended or subjective prompts. Don’t ask it to do anything creative or novel. It has its uses. Nowhere near what the snake oil salesmen would have you believe, and probably not worth the unsubsidized cost, but for now it has uses.
Yes and no. I find its terrible at solving more complex problems but its great at writing out tests for a function/view that covers every flow. My team went from having like 40% (shit) coverage to every PR having every case tested (inb4 they’re not good tests, they are good)
With that being said, fuck CEOs and fuck AI. At least you could (mostly) escape the blockchain hype
CEOs, embrace torches and pitchforks.
Copilot is shit.
Exactly, my company provides license for copilot and I use it, and while it has some highlights most of the time it actually is more a nuisance than help.
It especially annoys me because it hijacks autocomplete based on types with is own that frequently has subtle bugs, so now if I have it enabled I need to be on guard all the time. With the traditional autocomplete I could just trust it to be correct.
This is my experience. It saves a bit of typing sometimes but that’s probably cancelled out by the time spent correcting it, rewriting nonsense it produced, and reviewing my corworkers PRs that didn’t notice the nonsense.
Here’s where I’ll give it credit:
Doesn’t make up for the annoying-ass auto complete hijacking though. Stupid thing keeps making up non-existent functions and api’s and inserting them all over the place.
You have to put it in Ask mode so it doesn’t touch your code also ChatGPT models are free so if you want to ring up an AI bill use the Claude and Sonnet models.
You can turn off the copilot autocomplete in the ide and JUST use agent/edit/ask mode
i wonder if this the reason why its so bad on the phones, it autocomplete with words that arnt even close to what you are typing.
Turn off the autocomplete, it’s shit. Do use agent mode for targeted tasks that are easy but laborious. Don’t give open ended or subjective prompts. Don’t ask it to do anything creative or novel. It has its uses. Nowhere near what the snake oil salesmen would have you believe, and probably not worth the unsubsidized cost, but for now it has uses.
Yes and no. I find its terrible at solving more complex problems but its great at writing out tests for a function/view that covers every flow. My team went from having like 40% (shit) coverage to every PR having every case tested (inb4 they’re not good tests, they are good)
With that being said, fuck CEOs and fuck AI. At least you could (mostly) escape the blockchain hype