Agents and tools
The more technical posts. MCP is the standard that lets AI tools connect to your systems. What it doesn’t cover is orchestration: keeping multi-step work coordinated, recoverable and observable once several agents and tools are involved.
Claude Code is my example of an agent that works directly with files and tools, and it’s useful well beyond programming. The last post is the most speculative on the site: a longer view of how the economics of intelligence might change.
4 posts, in suggested reading order
The Model Context Protocol in plain English
MCP is a shared standard for connecting AI assistants to your systems. What it does, who governs it now, where it's risky, and where I'd start.
7 min read · Updated September 2026
The orchestration gap that MCP doesn't close
MCP gives agents access to your systems. Multi-step work across them needs state, recovery and coordination, and most of that is still yours to build.
5 min read · Updated September 2026
Claude Code does more than write code
Claude Code takes a task, works through files and tools, and checks its own results. That loop is useful well beyond programming. Where it stands in September 2026.
6 min read · Updated September 2026
Synthetic cognitive capitalism
A speculative argument that the bigger AI shift is economic. Know-how becomes capital, compute works like currency, and owning the context matters most.
7 min read · Updated September 2026