dhruv3006Amazon AI
hoshikarakitaridiaCoding is so easy
vibes = vibes-3
if (outage) {
fix()
}
I would love to know if this actually makes a difference though haha
I usually just ask it to recheck line by line for bugs with a thinking model and that’s pretty good at solving potential bugs, but they still need to be reviewed obviously. I only use open weight models though so they might be a bit inferior
driving_croonerWe are n weeks away for not needing developers anymore
ValmondPeople who share the size of a codechange as a mark of how effective ai coding agents are truly miasbthe point of code changes.
I wouldn’t say PR size is a bad metric, you usually just need yo read it the opposite of how sloppers do it, i.e. the most productive PRs are short and focused.
RustyNovaThat’s the beautiful thing about mandatory reviews. If I can’t understand the PR, I just veto it until it’s fixed.
I’m not saying it’s a good individual metric. In fact, applying individual metrics to developers (or most workers really), will only land you in Goodhart’s hell.
But as part of holistic operational health tracking, it’s a useful team level metric, as there is ample evidence that shorter PRs tend to result in less operational issues. And, of course, this is only valid if you don’t try to tie financial rewards to it, otherwise people will forget that shorter PRs is a proxy measure for how easy changes are to review and rollback.
KairuByteYeah, I’ve been guilty of coming up with “cute” solutions that are extremely optimized and concise, but you needed to take a hundred times as long to work through what was going on.
Usually I would put an explanation comment, but sometimes a less optimized solution is the better option for readability sake.
EthanYes, that’s what the comment said — smaller PRs are better.
And yet I experience it so often. That or “effort points” as the metric being used to determine who all stars are.
Either as a metric just encourages gaming of the system:
- Why write one line when I can write the same thing in 20?
- Why take this one effort point task I think will take three when I can just skip it and grab these one effort points I think will take 20 minutes?
I’ve been on teams that on the surface didn’t have these metrics matter, but the top effort points achiever got bonuses on the DL.
CorrectAliasIn college, on the first day of orientation, someone in my class bragged that they wrote 50,000 lines of code for a game that was similar to tic tac toe, emphasizing that he “wrote a lot of code”. A TA told him that it wasn’t a sign that his program was decent and that it really didn’t seem like it should take 50k lines of code to make something as simple as his game.
He dropped out after the first week of intro to programming.
That’s nothing, I wrote the code to return if the input is even or not in 1M lines of code.
Are you working with 20 bit integers? You’ll never get it down to a million lines like that…
There’s a million lines of If >4096, subtract 4096
Then a quick lookup case statement for whether the remaining number is even.
People who’ve never reviewed a PR be like. Wow green number big, is good.
Please tell me this subreddit is satire
mckeanRemoving shit and it still working perfectly the same is absolutely a goal everyone should have. Less code means less to maintain.
sqwDidn’t you hear? We’re going back to KLOC for measurement of f productivity.
If we’re going back to the 80s, do we at least get the company provided cocaine?
You reminded me of a story I recently read, where the author highlighted just how much awesome programming someone had done by describing how their hands were cramping up.
It’s like estimating how well an artist paints by looking at how much paint is on their clothes, or judging how good a cook is by how many cuts and burns they have. The actions that cause those things are incidental to the process, not central, and an excessive amount points to incompetence, not hard and skillful work.
softwaristBig Diff Energy
You know what’s funny? I use AI to develop software. However when I’m looking for libraries to do things if I see a CLAUDE.md file I have to look and see when it was added.
It’s like prewar steel.
It’s the difference between checking for questions in stack overflow and implementing solutions VS pasting every SO solution blindly until something works.
I do use autocomplete and ask plenty questions, sometimes even use an agent so it makes small changes that I then review and test, but I would never commit unchecked changes, and a claude.md implies that the AI is coding AND committing without supervision.
I can’t stress enough how different those scenarios are.
It’s not hypocritical. Because you use AI to code, you know how easy it is to just let the AI do it’s thing and not check it’s work. It’s almost like a sirens song. So you know the odds that a library that was coded with AI probably wasn’t checked by a human. That’s just called experience.
irelephant [he/him]
I love this sabotage. Has this worked against anyone that you know of yet?
Interesting. I’m unfamiliar with the purpose of that file as originally intended, but if I understand it right, it might be used to detect PRs with ai-made code. Tell Claude to write a particular string in every PR, and then reject all PRs with that string.
Jännät
‹Hexa«Back›ms themselves confirmed it, bragging 30% of windows updates are written by ai
Positive diff? Pfft, amateurs. If I ever see even a 1000 line PR I’m instantly rejecting and closing it. Learn to code, not generate bullshit.
tatterdemalion1000 lines isn’t that unreasonable for a PR. Commit size matters more.