What’s inside the PDF:
- The one-page job description: goal, inputs, tools, red lines, approval points, owner
- The first-month scorecard: one block per week, plus the month-end decision
- The go/no-go checklist covering logs, caps and the tested stop
A filled example of the hardest field, the red lines, for an enquiry-triage agent: never reply substantively to anything mentioning a complaint, a regulator or legal action, route those to the owner immediately; never quote fees; never contact anyone on the exclusions list. Three lines, written in five minutes, and they do more real governance than most AI policies, because an agent enforces what a memo only wishes.
For its first month the agent does its full job, but every output arrives as a draft for a person to approve, and nothing sends, saves or acts on its own. That produces a month of real evidence at zero action risk, and the scorecard turns the evidence into a decision: one block per week, then the month-end call. The wider method is the first-agent playbook. Fill the first page before anything is built, whether you are buying or commissioning; hand it to whoever builds, and hold the month-end decision to the numbers. The governance guide explains the six controls the checklist encodes.
This page and the template are for informational purposes only and do not constitute regulated financial advice, legal advice, or a compliance opinion. Adapt with a qualified professional where regulation applies.