A living notebook

Big ideas,
taken apart.

One topic at a time, explained from first principles — the naive question first, then the mechanism, then the parts that break. Words and pictures, no hand-waving.

The dissections
Nº 02 Latest
Agents & Interoperability

The Model Context Protocol

How do you plug any AI into any tool without building the glue a thousand times? MCP is the "USB-C port for AI" — the standard, its parts, and where it can be attacked.

~20 min · 6 chapters Read →
Nº 01
Agents & AI Security

Agent Skills, from first principles

A folder can teach an AI a new capability — and can just as easily hijack it. Progressive disclosure, the SKILL.md security surface, and how skill scanners fight back.

~18 min · 6 chapters Read →
Nº 03
Coming soon

More ideas, cracked open

Another complex topic, taken slowly apart with words and pictures.

In the works