I've been part of open source for most of my life. This year, the projects I love started closing their doors to outside contributors, and I don't blame them one bit, but it makes me sad.
A personal look at how AI is changing open source as I've known it.
And for programmers who are willing to use it carefully. My debugging time has gone to near-zero. “Here’s some log messages, and then a gdb backtrace of a crash. find out why”
10 minutes later, I have a code path for the error, and a fix. About 20% of the time the fix is garbage, but knowing what to fix makes it easy to know how to fix it.
My debugging time has gone to near-zero. “Here’s some log messages, and then a gdb backtrace of a crash. find out why”
Over in embedded land, setting it up so that the harness has the ability to flash and debug over JTAG/SWD (gdb interface to the entire microcontroller) is absolutely wild, especially for hard-to-reproduce bugs. I can set up a monitoring loop at the end of the day, have the bug I’m hunting for trigger at 2AM, and have a report waiting for me when I wake up in the morning that includes 7 different debug steps that were automatically flashed and tested.
Yes, it is mind boggling. A while back, there was a heisenbug in a distributed lighting controller I designed. I connected two boards to SWD, connected the serial bus to my scope, told Claude how to get to the scope over TCP, and let it loose. After a few hours it handed me a reproducible case and I was able to fix my hardware mistake that was glitching the serial port.
My general experience is that agents are so much better at everything else around software development other than writing code and I don’t understand why everyone focuses so much on it.
Debugging complex issues? Great.
Analyzing infra issues from logs? Huge timesaver.
Quick prototyping and throwaway scripts? Awesome.
Researching and summarizing stuff? Quite useful, although a bit unreliable.
Writing code? Mostly garbage code that requires a rewrite to be maintainable.
And for programmers who are willing to use it carefully. My debugging time has gone to near-zero. “Here’s some log messages, and then a gdb backtrace of a crash. find out why”
10 minutes later, I have a code path for the error, and a fix. About 20% of the time the fix is garbage, but knowing what to fix makes it easy to know how to fix it.
score 1 · canonical
Over in embedded land, setting it up so that the harness has the ability to flash and debug over JTAG/SWD (gdb interface to the entire microcontroller) is absolutely wild, especially for hard-to-reproduce bugs. I can set up a monitoring loop at the end of the day, have the bug I’m hunting for trigger at 2AM, and have a report waiting for me when I wake up in the morning that includes 7 different debug steps that were automatically flashed and tested.
score 1 · canonical
Yes, it is mind boggling. A while back, there was a heisenbug in a distributed lighting controller I designed. I connected two boards to SWD, connected the serial bus to my scope, told Claude how to get to the scope over TCP, and let it loose. After a few hours it handed me a reproducible case and I was able to fix my hardware mistake that was glitching the serial port.
score 1 · canonical
My general experience is that agents are so much better at everything else around software development other than writing code and I don’t understand why everyone focuses so much on it.
score 2 · canonical
LLMs have been trained on the Internet. And for anyone who’s seen the Internet, it’s full of crap.
score 1 · canonical
deleted by creator