Post

sarcastic parrot
@monkchips.com

Throwing away code as part of building. "Falsework", like building architecture - the falsework is the stuff you build when putting up a bridge, that you later remove leaving the actual structure you were building for. Interesting idea at this @honeycomb.io O11y day.

1 repost18 likes

Chris Blume
@programmax.net

Aye, the Phoenix Architecture coined by Chad Fowler (a spin on Martin Fowler's Phoenix Servers). I'm actually about to test this now on one of my own projects. Although, I wouldn't describe it as "later removed". I would describe it more like code gen. I still check-in and keep the generated code.

sarcastic parrot
@monkchips.com

"Money is real, even in a fully AI-pilled organisation." - Brian Scanlan at fin. Fascinating talk this, at @honeycomb.com observability day. Lots of progressive delivery in how fin rolls out and manages its developer platform. "Don't get an agent to fix something every time, just use a script "

sarcastic parrot
@monkchips.com

Only small pull requests are accepted. Must be 150 lines of code or less. This is still the policy at fin, even with agents writing the code. Pull requests with thousands of lines of code will never be safe, no matter how much you review it. @honeycomb.com Observability Day. Ship small, ship safe.

sarcastic parrot
@monkchips.com

Fascinating talk by Jana Chen, an SRE at baseten, an inferencing platform, about using Claude as an assistant in bringing systems back after a major incident. When do you need to pay attention, if "attention is all you need"? The human SRE still needs to be very attentive to "fixes".

devtocash.bsky.social
@devtocash.bsky.social

The attention cost shows up as anchoring: the agent's first confident hypothesis narrows what the responder bothers to check. What has worked: every claim carries its raw evidence (the exact PromQL, log line, kubectl output) so verifying takes seconds, and every write stays behind a human approval.