@dklineii “Eat the chicken. Spit out the bones.”
A helpful adage for processing feedback.
“Ask for seconds.”
Closing the initial loop while expressing it was valuable to you = inviting more feedback. And so the team gets better and better.
@kieranklaassen@nbaschez Feature diversity probably determines how this plays out. Better fit for a lot of interwoven code. It sounds like a stacked branches flow but without intermediate review of the stack. But why not review per branch at the end of the day instead of a one big diff?
We're dealing with a major malicious attack on @rubygems right now. Signups are paused for the time being.
Hundreds of packages involved - mostly targeting us, but some carrying exploits. The team has been on this for hours. More details to follow once we're through it.
#ruby
@GergelyOrosz SRE looks silver now, but for most of the time, it probably felt more like lead. Constant effort and attention to push reliability forward over two decades. Throwback to this classic: https://t.co/3FMsvaG28h
@palkan_tula Make it work. Make it right. Make it fast.
I like to organize along those three iterations. Forcing a solution early misses the learnings of a good process.
@dieworkwear When you are new to something, you start testing previous behavior and ideas to decide is this something to keep or discard. Zeal is a social expression of internal redefinition. “Once I was blind, but now I see.”
Matt has been compiling the best researched Ruby AI newsletter on the planet.
If you're not reading this you're really missing out.
I'm in this space and I always learn so much from it!
Thanks Matt.
Finally seem to have sorted this out. If you're using coding agents on macOS, there's an easy to miss source of failures.
macOS ships BSD versions of standard tools like sed, grep, find, and awk. Most examples online (training data) assume GNU/Linux versions. These are, surprise, not the same.
The commands run, just with slightly different behavior.
Installing the GNU tools removes a potential mismatch and aligns your local environment with an agent's likely assumptions. Harnesses and training will evolve to resolve the issue, but this worked for me today. Sharing it here hoping it works for you too.
The simplest fix is to install the GNU versions of these tools and put them earlier in your PATH.
Install GNU utils
`$: brew install coreutils findutils gnu-sed grep gawk`
Add to PATH in your shell (.bashrc, .zshrc, etc)
```
for pkg in coreutils findutils gnu-sed grep gawk; do
gnubin="$(brew --prefix)/opt/$pkg/libexec/gnubin"
[ -d "$gnubin" ] && export PATH="$gnubin:$PATH"
done
```
Generally satisfied with results when I include “…referring to https://t.co/KkdobsOua2 as needed” or a particular concept specified within, like scoring criteria to prevent scope creep. Preloaded, it only seems to increase the probability of decisions based on the content, but to a widely varying degree. Maybe it has a hard time determining how important the various rules are to the current task.
@simonw Either way, it is still vibes and taste, but the leadership is swapped. Vibe coding lets the AI lead, a wave to be surfed. Vibe engineering leads the AI, paving the road to be driven.
@steveruizok Basecamp is a great fit for this. Projects get a bit of structure, accountability and visibility, but not so much it becomes someone’s job to organize.