Post #4459411

2026-08-09 04:59 UTC

@daisy@cloudisland.nz @mcc@mastodon.social the most frustrating thing to explain about this whole mess is that people believe, they believe with their whole heart, that they are built different. that when _they_ submit slop, they have verified their understanding and carefully vetted every line of code, that they "take responsibility" for it. and sometimes they do! but you know, they do it like 30 or 40 times, and then the 41st time it's probably not such a big deal, they just slip a little bit, the LLM is *usually* right…

Replies (3)

@daisy@cloudisland.nz 2026-08-09 05:04

@glyph@mastodon.social @mcc@mastodon.social Yeah. My personal experience is that in simpler, more interactive workflows, the LLM is confidently and sanctimoniously wrong often enough that I don’t feel I can trust it and stake my professional reputation on its output without independently verifying *everything*. And that makes me distrust more “advanced” workflows that involve less direct supervision. And while I understand that “agentic” workflows include “harnesses” and “guardrails” that attempt to automate the process of verification, I don’t necessarily trust my own ability to write those, and I think that the assumption that it is possible to exhaustively enumerate upfront every possible failure case feels like an exercise in hubris.

Open ##4459410
@glyph@mastodon.social 2026-08-09 05:00

@daisy@cloudisland.nz @mcc@mastodon.social they still defensively add comments to every PR, roll their eyes in exasperation, "it's just a tool" they say, "sure, more JUNIOR folks can't handle it, but I can, I know what I'm doing" and they just slowly don't bother to check and a few hundred PRs and dozens of exasperated self-rationalizations later, they're freehand vibe-coding their way through massive production configuration changes, slamming the "merge without checks" button because it's been fine *so far*

Open ##4490589
@elseweather@mastodon.social 2026-08-09 05:13

@glyph@mastodon.social @daisy@cloudisland.nz @mcc@mastodon.social I have to* use this stuff to some degree for work. And I honestly try to review the PRs I put out this way. But I am always shocked by the stuff the next reviewer catches. It's always mistakes that I would Just Not make if I were building up the code myself

Open ##4490592