notes

AGENTS.md

@@ -41,6 +41,8 @@ we use  `yx` for planning run yx --help to see options. Invoke when the user wan
 
 `yx list` - shows open yaks
 `yx add` - adds yaks
+`yx context "yak name"` - adds human-readable description (pipe via stdin)
+`yx show "yak name"` - shows yak details including context
 
 # Building
 

app/priv/blog/engineering/2026/05-06-breaking-work-down-with-yaks.md

@@ -0,0 +1,22 @@
+%{
+  title: "Breaking work down with Yaks",
+  author: "Willem van den Ende",
+  tags: ~w(),
+  description: "",
+  published: false
+}
+---
+
+Chris Parsons asked me what it took to get started with Pi. To be honest, I can't tell really, even when I think I work slowly, many things happen at once. Yesterday I did what felt like a tiny amount of work, while doing other things than development. When I sat down this morning to take notes,  I already came out at multiple blogposts. I guess I am going to draft them all, and see what makes for a useful sequence.
+
+Yesterday I loaded up the dense model, thinking "this can go slowly, I need to do other things. So if Pi does its' thing while I get some code out, that would be a win". And it was. Except now that it almost runs fast enough, I had some interactions, did some thinking and set up Pi on a second, eh, third machine.
+
+At least setting it up again helps me answer Chris' question in part. Simplicity is when there is nothing left to take away. So I start from a blank canvas, and pull in models, stored prompts (e.g. commands, agent skills) as needed. I saw a message from *the agentic dev team person* yesterday saying that his commands etc are per project. 
+
+I had a relatively minimal Claude Code setup already. Claude is a bigger harness. 
+
+# What
+
+# So What
+
+# Now What