I guess things keep getting more features and complexity but all of it is 80% done because it's "shippable" and updates are cheap.

This is most visibile for me on Windows/MacOS and complex web apps (e.g. GitHub and GitLab, including consoles of hyperscalers), where 80% of "normal" things work, then you need the last 20% and it's always not working as documented, only half-working, or just outright broken, and you need to find "temporary" workarounds that stay in place for years.

I feel this is being amplified by AI: tests come last (if at all) and are still written by LLMs, nobody really looks at them anymore, green pipeline checkmarks mean less than they did in the past.