Architecture & Philosophy

How 2,600 lines of discipline changed what AI agents could become. The origin story of Amplifier's kernel philosophy — mechanism not policy, the crystal lattice metaphor, and why the center stays still so the edges can move fast.

The composition model that makes AI agents as simple as YAML and as powerful as your imagination. How the thin bundle pattern, behaviors, and LEGO-like composition create infinite possibility from minimal declarations.

What happens when you put an AI agent framework inside a single HTML file. No servers, no API keys, no accounts — just open a file and you have a complete AI assistant running locally on your GPU via WebLLM and Pyodide.

40+ specialist agents, one conversation, zero complexity. How the context sink pattern, parallel dispatch, and delegation trees create the illusion of simplicity over a team of coordinated experts.

A design case study on building an AI specialist that selects and narrates rather than summarizes, leveraging narrative cognition.

An architectural deep-dive into amplifier-core. The actual file map with line counts, protocol signatures, coordinator mount points, session lifecycle, event taxonomy, hook result actions, and what's deliberately NOT in the kernel.

How markdown files become AI agent systems. The architecture of amplifier-foundation — bundle data structures, compose() merge semantics, the prepare step, namespace resolution, spawn mechanics, and the three-layer architecture.

How cognitive task mismatch breaks prompt-level format enforcement — and why the fix is architectural, not instructional.

A2A communication and multi-directory architecture for maintaining context across distributed agent networks.