Blog
Exploring agent engineering, cognitive architectures, and the future of AI.
New to seed agents? Start Here for a five-minute overview before diving into the essays.
From Apprentice to Architect
A seed agent is not a product. It is a thing that gets installed, then grows under the operator’s care. Jobs mature through four stages; behavioral controls migrate from voice to hook to plugin. The brain stops growing in size but never stops learning. Start at Part 1: Apprentice to Architect Foundation.
Start the series →The Plugin Kit
A plugin is a cell with internal organs. CLAUDE.md, hooks, scripts, two voice surfaces, hidden state, a documentation layer with a historian, per-plugin subagents, tests, and a lock ceremony that refuses to commit when tests fail. The kit gives the brain the capacity to grow. Start at Part 1: Plugin Kit Foundation.
Start the series →The Markov Phasic Brain
The seed agent’s cognitive work happens in five phases plus one cognitive organ — observe, plan, execute, verify, condense. Nine short essays walk through the phasic layer one phase at a time, then close on the backward multiplier and the off-cycle GMODE lane. Start at Part 1: Phasic Foundation.
Start the series →The Always-On Digital Cortex
The seed agent doesn’t have memory — it has a bus. Nine short essays walk through the always-on plugin layer one plugin at a time, then build up the CLAUDE.md hierarchy, the historian-ratchet ceremony, and the customization guardrail that gates substrate edits. Start at Part 1: The Two-Layer Foundation.
Start the series →The Folder Is Alive
Add a .claude/ directory to any folder, and that folder becomes an agent. Twenty such folders, taught one job at a time, become a personal cognitive workforce — your liver, but for paperwork.
The Language of Agents
Every industry builds a wall of jargon. AI built one faster than any industry in history. This post hands you the keys — from LLMs to seed agents, one read, every word that matters.
Read article →Your Brain Was Never Built for This
Your skull holds a brain optimized for spotting snakes in tall grass. You are asking it to run a business in a civilization of eight billion people. CLI agents are the digital cognitive organs your biology never built.
Read article →We Could Have Had AGI By Now
The building blocks for durable, self-improving agents have been available for months. If we had scaled architecture instead of scaling only the model, we could have had AGI-like autonomy already.
Read article →LLMs Are Not the Agents
Most people point at the model and call it "the agent." That is the root misunderstanding. The model is the engine. The agent is the engine plus a local brain — a filesystem that gives the model memory, structure, and identity.
Read article →