Now

Updated September 2026 · This is a now page — a snapshot of current focus, refreshed when the focus changes.

Building

Eleven reference builds now anchor the site: an agent memory layer; an evaluation and observability harness; Sutra Feed Guard for deterministic external-feed change control; agent privacy as a data-flow architecture; an LLM gateway control plane; an exception queue before happy-path automation; policy-as-code for enterprise agent governance; and four additional production-seam builds covering routing, durable execution, document extraction, and sandboxing.

Focusing

This site is organized around one claim — designing agentic AI systems that survive production — and the garden now argues it across agent architectures, data foundations, and operating practice. The breadth (architecture, data platforms, product, design, teaching) has settled into its proper role: the foundation under that claim, not a competing set of interests.

Writing

The garden is at 267 notes and planting daily. The newest notes separate security detection from containment authority, keep read-only MCP knowledge out of executable tools, require outside opponents before self-play production claims, and qualify multilingual agents with per-language action evals.

---

Want to talk about any of this? Email me.