The Playbook Ratchet: Why Process Improvement No Longer Ends
Published on: 2026/10/08
A playbook only encodes the attacks someone already ran. It raises the floor and says nothing about the ceiling, so the human job converged to three things.
Ask for Attacks, Not Feedback
Published on: 2026/10/06
AI has two evaluation modes, and only one is biased toward telling you what you want to hear. Stop asking for review; ask it to break a specific commitment.
ask for attacks
Irreversibility: The Only Sane Way to Allocate Executive Attention
Published on: 2026/10/01
AI work is disposable and regenerable. What compounds are the commitments you can't take back. Route senior judgment by one-way doors, not by cost.
Why \"Human Review\" Fails, and What Adversarial Oversight Looks Like
Published on: 2026/09/29
When the machine is right 95% of the time, the reviewer becomes a rubber stamp. If the sign-off can't say what they tried to break, it was never oversight.
human review