Tech Shall Not Kill

*Tech Shall Not Kill: Building the Consequence Layer for AI Agents*

Your AI agent followed its instructions. It still issued the $640 credit, changed the appointment, and moved the due date-because nothing outside the model was able to say no.

Most organizations governing AI agents write clear rules, careful prompts, and thorough audit trails. Then they discover, usually after something goes wrong, that a well-written instruction is only a note on the cash drawer. It does not lock the drawer. *Tech Shall Not Kill* is about the missing machinery: the independent Consequence Layer that sits between what an agent proposes and what actually happens in the world.

Through the story of Harborline Services-a regional operations company that adopted agents one department at a time-Edward Fassio shows how ordinary controls (permissions, budgets, approvals, emergency stops, and durable records) can be arranged so an agent cannot simply route around them. The book moves from the first unpleasant surprise through the practical design of gates, sponsors of record, reversible versus irreversible actions, and the economics of enforceable trust.

Written for the same table-a high-school senior, a manager, and an engineer-it explains technical ideas once, in plain language, and ends every chapter with a field note you can use the same week. There is no code. There is a clear distinction between the sign and the fence, between the camera and the gate, and between a track record of good behavior and actual authority.

This is the fourth book in The Agentic Frontier Series. Earlier volumes asked what must remain true and what remains for people when agents buy, sell, and decide. This one asks the operational question those books left open: once a boundary has been agreed, what actually holds it?

If you are putting agents into real workflows, this book shows how to make "shall not" mean something the systems will enforce.