Post #2574540

2026-05-13 22:46 UTC

IMO it&#39;s going to get *much* worse before it gets better. The cost of discovery has gone from “months of expert time” to maybe “hours of expert time&quot; with this new wave of tools; and most of the experts *might* not even have access to the “best&quot; tools, yet. (citation needed on that one, but… there are power-multiplier tools, certainly, for people who use them). I think we&#39;ve barely scratched the surface on software that is exploitable (spoiler: nearly all of it). @ello https://void.ello.tech/@ello/statuses/01KRHFM1ESH4HKK73XRY38MAZ5

Replies (1)

@ello@void.ello.tech 2026-05-14 05:31

@sean@scoat.es there was a poll on here a bit ago about whether agents would be the future of software engineering, and the results were something like 75% no. we're mostly the anti-LLM set here, and that's good and right, but i think people who don't use them don't realize how effective these machines are when you build deterministic verification systems into the process and apply some rigor. in software, we already do that. is the vibe-coded todo-list app going to fall over after it exposes your credentials? yes. the postgres extension i built with full unit and integration test coverage and multiple rounds of reviews? no. or at least, it's not more likely to fall over than anything i built by hand. anyway the opposition can't be they don't work.

Open ##2864246