Technical posts on building runtime governance for autonomous agents. Philosophical posts on what we're really building, and why this layer has to exist.
A viral post says the atom is light that got stuck. Physics half-agrees. The Arizal half-agrees. The Alter Rebbe says no — and his reason is the case for alignment as architecture, not walls.
Chana did not say she asked for the boy. She said she borrowed him. A ruling in four words about what a child is — and the one clause in the law of bailment where the borrower owes nothing at all.
A response to Rabbi Sholom Kesselman's "Chassidus Misapplied": on ohr and keili, a letter of the Rebbe on bending, and the trajectory of his last decade.
On the maamar HaShkifa MiMe'on Kodshecha, Chai Elul 5727: the flaw that reaches the essence, and the repair that alters not one thing it touches.
The analytical companion: the snow/wool pair taken apart along its axes — traction, direction, operation, joining — with the anatomy table and the transmission-versus-relocation figure.
Most agent failures are recoverable. Some are not. The ones that aren't are the ones the field is least equipped to talk about — and they happen earlier in production than anyone is admitting. A taxonomy, and a proposal.
A short and very practical case for keeping the safety classifier outside the model that produced the action. Includes benchmarks.
The constraint is not the cost of the system. It is the system. On the difference between alignment as a wrapper and alignment as architecture — and why the second is the only one that holds at runtime.