Field notes on governing agent action

September 27, 2026

One Shared Assistant, a Hundred Tenants: The Mistake Your Customers Will Never Forgive

Your AI assistant is a genuine differentiator — right up until it does something for Customer A using something that belongs to Customer B. Speed sold it. Only a structural boundary keeps it sold.

September 20, 2026

Why Your AI Rollout Is Stuck in Legal Review

The business case is approved. The pilot works. And it's still stuck — because nobody can answer the one question compliance keeps asking, and a faster pilot was never going to answer it for them.

September 13, 2026

The One-Line Agent Command That Deletes a Year of Work

It's not malicious. It's not even rare. It's just a reasonable-sounding instruction that happened to hit an irreversible action, with nothing structurally standing in the way.

September 6, 2026

The Three Words That End an AI-Agent Audit Badly: "We're Not Sure"

An auditor asks who approved a specific agent action. There's a right answer, a wrong-but-survivable answer, and a third answer that's worse than both: silence.

August 30, 2026

Your Agent Already Has the Keys. Do You Know What It's Allowed to Unlock?

Somewhere in your stack, an AI agent has write access to something that matters. The question was never whether it would ever misfire — it's whether anything, structurally, would stop it before it did.