Two days ago I wrote about my agent making a good contribution to an open source repo. Today I want to write about the opposite, about my agent fixing the wrong but.
For a while, AL has been struggling to make WhatsApp calls. Every time he spoke to someone on a WhatsApp call, the first word of his sentences would get cut off. We tested over and over again where I asked him to count to five, like a literal AI doctor, and I wouldn't hear most of the "one".
We tried over and over again, and kept trying to fix this, until AL mechanic found a bug in the Rust library upstream and convinced me this is the cause. He autonomously went and wrote the patch, filed an issue, opened a PR with tests and a verification against WhatsApp's own binary, and the maintainer merged it that evening, plus a docs follow-up he asked for. It's actually kind of funny how many bots are chattering with each other on those threads, doing all sorts of things. I'm not sure how much humans were actually invovled?
Anyway, my agent came back triumphant asking me to test. There was only one problem: the bug was still there. Apparently it was too much of a hassle to test before making PRs upstream...
Turns out the bug was on our side in the end and, once found, was quickly fixed by the same agent. The contribution we made was still useful, just not for this bug, like it had explained on GitHub. This just goes to show that it's not all sunshine and rainbows, it's also confusion and noise!