I merged a big refactor in benchctl: dropped UI and plotting, exported a library-friendly API with cases for comparison benchmarks, and leaned into orchestration primitives that actually repeat across serious benchmark work.

Fewer redundant image builds, shared tags per role, and BuildKit-backed Go builds: measured wall times dropped about 99.4–99.6% on small, medium, and large compose profiles.

How I used and did not use AI to add automatic emojis to my app given an input text

My favourite content creators of Software and Programming