AI
Jun 10, 2026
•
5 min read
How a 22-person team runs on roughly 60 governed AI agents, and the map to every article in this series.
Jun 9, 2026
We could not prove our AI's ROI. Almost nobody can. Here is what we measure instead.
Jun 8, 2026
Every AI stack has handoffs. Ours is built so context compounds instead of leaking.
Jun 7, 2026
Anyone can rent the model. The compounding comes from the context you own.
Jun 2, 2026
10 min read
Four ways the team meets the machine: personas, alerts, memory, and a human in the loop.
May 31, 2026
9 min read
The leverage is not in the model. It is in what the model can read and act on.
May 28, 2026
12 min read
Cron-driven operating loops, CI/CD that ships its own code, and the telemetry that lets the system run itself.
May 26, 2026
Persistent memory, scheduled task queues, and a fleet of containers that never lose business context.
May 24, 2026
11 min read
The architecture, security model, and self-improvement loop behind an assistant the whole team shares.
May 21, 2026
Keeping 60 agents accountable: a registry where every agent is a row, and a propose, approve, apply loop for every change.
May 19, 2026
6 min read
Why we turned Salesforce off after an eight-week cutover, and what we would do differently.
May 17, 2026
7 min read
How Notion AI, Claude, and NanoClaw divide the labor, and the MCP layer that connects them.
May 14, 2026
The schemas, relations, and design patterns that turn a workspace into an agent-ready substrate.
May 12, 2026
Before AI could run the company, we had to rebuild the company into something a machine could read.
