love the title bro „stars don’t pay bills“ 😄

Hard to disagree. Personally, to make it bit more nuanced: it’s the „sub-tasks“ which I’m sometimes not good at and where LLMs massively help. Example is is writing (style, grammar, flow) which is hard for
me but foundational to things I‘m good (and now becoming better) at

…creating a feedback loop based on recording transcripts and doc comments (works perfectly with LLMs) which then helps to reduce any „bloat“ - incl. updating rules for next iterations.

(more general since I also use it for writing) I aggressively use rules and „personas“ to quickly iterate and keep the machines in check. sometimes I don’t care about bloat if it’s throw away code (eg Python to generate excel stuff which is the artifact). in writing, we have lots of reviews…

I‘ve become a huge advocate internally and would make for a good salesman at one of these AI companies 😆

I think *most* errors should default to returning a "safe" error message that doesn't include any raw input data or sensitive data about the host system, and then add an optional field/method that has more detailed info that might be sensitive (likely only used during debugging).

:)

Yup, that’s the canonical I also learned from @edjgeek.bsky.social :)

@mreferre.bsky.social can certainly help and connect :)

dead link :/

😆

Let’s settle the debate and use foopkg.WithConfig(c cfg) 😉

still learning, my friend, still learning :)

For people who felt uncomfortable, needed time to adjust or were afraid to make mistakes eg by asking questions, was there anything you did proactively?