<!channel> we are being rate throttled by docker f...
# software-contributors
m
<!channel> we are being rate throttled by docker for pulling to many images in our rubber. Anyone available to help us setup better caching with ECR so we don't pull so much from docker? I already created the ECR repo but since we are pulling images as part of third party GitHub actions, I'm not sure how to get them to use the ECR repo first.
s
Replying only to let you know that I saw this. Unfortunately I have zero experience in this area, and I'm pretty certain your Google and chatgpt skills are at least as good as mine
I do have a little free time this morning if you'd like to pair on it just to get another set of eyes. But aside from that, I'm hoping you find someone who can help more!
m
thank you for the reply!
have you tried codex? itโ€™s not as good as devin but its free
(with your chatgpt plus sub)
s
Nope, also I don't have chatgpt plus
m
gasp
s
I had it for a bit while my company was covering it, then they stopped, and I cancelled out of spite. You want me to pay to be more efficient for you?
Between cursor, Claude code, and chatgpt anywya I have lots to learn for now
m
cursor isnโ€™t worth learning if you have someone paying your claude code subscription
also, my 2 month projjection is that codex will surge pass claude code
i think they are heading in a better direction
s
Is Claude code useful in conjunction with an ide? I've used it maybe once (and really do need to go back) but it's so separate from the code I'm used to staring at, I'm really not sure how to make it work with how I work
m
claude code has an IDE plugin which i find helpful bc it lets you view the diff and approve it easier
locally, your best option is claude code
remotely, your best options (assuming cost is not a factor) are in this order -> devin, claude-code, codex, then cursor background agent
s
Haha my first time I asked it to do something, it was on a parent repo with submodules, it told me it wrote a bunch of code but I was never able to actually see it ๐Ÿ˜ญ
m
you should see it all in git diff, it might have had an issue writing
before it writes code for big repos, you need to have it right documentation
so it knows what does what and why
s
Git diff showed me different commit hashes for the submodules, but within the submodules there were no changes. I'm going to have to try it again for a simpler repo structure
m
you can ask it to write documentation that you have to approve for each submodule
s
So is the workflow to have it write code, and then use git diff to see that code? (And of course unit tests etc etc)
m
no, claude code (unless you turn on yolo mode) presents each change to you for approval as it works
๐Ÿ‘ 1
b
@mysterious-tomato-10057 we're stuck with gpt-pro, you're making me jealous lolsob How do you know all the pros and cons of each option? I tried researching this stuff but never got this detailed
m
i've spent many hours on each
my full time job is to write code, so i am up to date on all the tools that help me do my job ๐Ÿ˜‰
b
You don't have privacy issues?
or are you paying for the plus version of everything?
m
for frum.finance i get to make the decisions, so we use devin, claude-code, codex, and cursor. hubspot has licenses for claude-code, chatgpt, and cursor, so i get to use them all there.
when i was at amazon where we didn't have those licenses, i would use them indirectly, and not give them code access
though amazon had an in house version of aider basically which bake then was before claude code existed so it was pretty powerful for CLI based LLM
b
cool stuff
m
"You won't lose your job to AI. You'll lose it to someone using AI."
๐Ÿ’ฏ 1
๐ŸŽฏ 1
b
Very true ๐Ÿ˜ญ. Any workshop recommendations? I know you went to one recently
m
but besides for that it's just using the tools that will teach you
no workshop can give you experience
need to develop a feeling
๐Ÿ‘ 1