Implementation-focused research notes on AI governance, agentic systems, and regulated deployment. Citable, evidence-led, distinct from ordinary blog posts.
Claude Code subagents used 2.6x to 5.9x the tokens of working sequentially, and were never faster. We measured the fan-out tax and the fixes.
We measured what Claude Code and OpenCode spend before reading your prompt, then added instruction files, MCP servers and subagents to the bill.
If a Tuesday conformity assessment says little about Wednesday, the static model breaks down for orchestrated AI. A continuous-conformity proposal.