The Bicameral Architecture for AI Research Agents
An architectural pattern for AI research agents that separates reasoning from arithmetic, so a language model can decide what to look up and how to explain it, but never touches a calculation.
Knowledge and wisdom on craft, attention, and creative process. Written for those who find meaning in focus and purpose.
An architectural pattern for AI research agents that separates reasoning from arithmetic, so a language model can decide what to look up and how to explain it, but never touches a calculation.
Claude Code's VS Code extension waits until you stop typing before surfacing a permission dialog, then restores your input exactly as you left it. A snap note on what that small decision reveals about interruption UX and user state preservation.






A useful prototype answers one risky question clearly. Polish applied before that question is answered makes an idea comfortable to approve while leaving its hardest problem untouched.





A practical walkthrough of the merge, reduce, pivot, and column math operations that together replace almost every for loop in a production data processing pipeline, with worked examples and the reasoning behind each.

