Agent Runbook

Agent Runbook

Operational writing about running AI agents in production — what they cost, what they are allowed to touch, and how you find out what they did.

Three articles so far. Every one of them states its method in the first paragraph: what was measured, what was modelled, and what was not tested. If a number came from a vendor's pricing page rather than our own instrumentation, it says so at the top rather than in the small print.

Articles

What this is

A small publication, written by the team at Agency Agents, about the unglamorous half of agent engineering. Not model releases, not benchmarks, not predictions. Cost arithmetic you can redo with your own numbers, permission checklists you can run before connecting a server, and tool comparisons that admit where the published figures disagree with each other.

The articles are deliberately few. In a niche where anyone can generate a hundred posts a week, volume is not a signal — so we would rather publish three articles with their assumptions exposed than thirty that restate documentation.

Who writes this

Agency Agents builds and operates agent fleets, and has built a small number of skill packs that encode the same practices these articles describe. That page describes them, and is explicit that they are not currently available to buy.

This site is a working build. It is not indexed by search engines, and nothing here is an announcement.