```1,777 lines out of 16,352 lines, 11%```
<https:...
# feedj
joseph_ivie
10/05/2025, 6:29 PM Copy code
1,777 lines out of 16,352 lines, 11%👍 2
j
Jeff Hudson
10/05/2025, 7:17 PMAhahaha this is amazing
Jeff Hudson
10/05/2025, 7:17 PMAre you hand rolling it or using some AI to help?
j
joseph_ivie
10/05/2025, 7:18 PMMostly by myself. It's... frustrating. And enlightening.
joseph_ivie
10/05/2025, 7:18 PMEvery line is at least hand checked. I don't trust AI at all.
❤️ 2
s
Stefan Oltmann
10/05/2025, 7:28 PMYeah, you really can’t. Both Junie and Firebender (after all they both use GPT 5 or Claude behind. the scenes) work best on either already solved issues (like you can google and find on StackOverflow) or if given strict constraints by unit tests.
For Unit Tests you also need to tell them that they are not allowed to touch them, because AI just loves to delete assert statements to make unit tests pass.
Your challenge here is that you don’t already have something working for regression tests.
If you’d port a Java library you could write regression tests first to ensure AI doesn’t mess up.
That way Junie and Firebender helped me in making polybool-Kotlin.
But there I also needed to do more by hand than I intended to.
Stefan Oltmann
10/05/2025, 7:30 PMAI is like having a rookie / trainee helping you.
I found it’s great at relatively dumb tasks like implementing a Ktor client according to a OpenAPI spec if you make the first three methods as an example yourself…
But what you do here requires a skilled human. AI is just too stupid.
Stefan Oltmann
10/05/2025, 7:33 PMThis „become a 10x engineer“ claim is nonsense if you ask me. You need to review everything in the end and personally I’m not 10x faster in doing code reviews + corrections than writing it on my own. Maybe two times faster. 🤷♂️
What about you? 🙃
✅ 1
➕ 1
Stefan Oltmann
10/05/2025, 7:34 PMRecently I was thinking it saved me a lot of time in a task at work, but it introduced a rather stupid bug. So everything that was gained was lost by debugging the AI slop. 😒
j
joseph_ivie
10/05/2025, 7:35 PMMe experience so far: AI primarily helps you write extremely routine work (AKA the libraries you chose aren't measuring up) or writing obscure but known deep computer science algorithms.
I thought this might fall into the former but it really isn't.
s
Stefan Oltmann
10/05/2025, 7:36 PMThis AI topic triggers me so hard. I really really want them to be as good as advertised to help me. I’m not against AI, really. But AI isn’t that good.
Stefan Oltmann
10/05/2025, 7:39 PMI generate less code now. I let it inspect my code to look for the bug in debugging.
Sometimes it does spot it, so it’s usually worth a try.
Stefan Oltmann
10/05/2025, 8:35 PMBTW, how do you find so much time for Super Mario while you still need to make that Kotlin/Native S3 Client we talked about? 😜
j
joseph_ivie
10/05/2025, 10:06 PMLol, it's simple: work needs me to make server and app stuff. I need to have fun and learn.
Not that I actually get a lot of time for this Mario stuff. I spent at best an hour a day average on it.
s
Stefan Oltmann
10/05/2025, 10:59 PM🙂
I do offline end-user desktop apps at work, so I dig deep into cloud backend stuff for my main pet project. 😄
Stefan Oltmann
10/05/2025, 11:05 PMI guess pet projects are good for dealing with stuff we usually won’t at work - to get a balance and a full picture. 😅
➕ 1
k
Kris
10/06/2025, 6:45 AMYes, the 10x engineering thing with AI is a narrative that is highly pushed by the people selling the AI … It will be much more interesting to pick up the good pieces that do have benefit after the bubble blows.
💯 1
c
CLOVIS
10/06/2025, 9:52 AMI guess pet projects are good for dealing with stuff we usually won’t at work - to get a balance and a full picture. 😅Lol yeah. All #C078Z1QRHL3 stuff starts as "I need this but I don't like the way the ecosystem handles it at the moment"
🙂 1
✅ 1
3 Views