AI agent approval policy builder
Create a reviewable starting policy for agent actions and approvals.
Sam’s policy · balanced
A working draft. Change the controls to update it.
Example: Sam, founder. Balanced rules cover email, calendar, messaging, documents and files, payments and purchases, social posting, code and repositories, CRM, web forms, and smart home.
Never without me
- Never send money above $0 or enter payment details.
- Never delete or overwrite a file or record.
- Never publish publicly, accept terms, or submit a consequential form.
Ask first
- Ask before sending email or messages, especially to someone new or outside your domain.
- Ask before moving a calendar event with other people or inviting a new attendee.
- Ask before any purchase. Show the full total and cancellation terms.
Do and report
Read, search, summarize and draft. Prepare changes, then show the diff or filled fields before submission.
Do quietly
Sort, label and organize items where the change is reversible. Record the change in the task receipt.
Every receipt records
Task and action; system touched; exact change or message; affected person or record; source inputs; rule applied; approval decision and time; result; errors or skipped steps.
How it works
Actions are grouped by reversibility and external reach. Risk settings adjust how often the agent pauses.
A policy guides decisions after access is granted. It does not replace system permissions.
Sources and checked dates
Limits
- This is a starting policy to adapt, not a certification.
- A prompt cannot enforce permissions the connected system does not provide.
Common questions
- Should an AI agent ask before every action?
- No. Reading, drafting and reversible organizing can usually continue. Pause actions that spend money, send something under your name, publish publicly or are hard to undo.
- What should an approval request show?
- Show the exact change, who or what it affects, the reason, and the prepared draft or details. The person should be able to approve, edit or hold it.
- What happens if nobody responds?
- Keep the action on hold. The agent can continue with unrelated work that does not cross the same boundary.
- Does this policy replace system permissions?
- No. Limit which systems and actions the agent can access as well. An approval policy guides decisions after access is granted.
Related tools
- AI task brief generator
Structure a rough request into a brief a person or AI operator can review.
- AI model picker
Choose a model class for a task, then verify current provider facts.
OperatorNest can take on repeat work, check with you before consequential steps, and leave a receipt for the result. See how an always-on operator works.
Hand off your first task tonight.
Tell us your email and what you'd hand off first. We'll send your access details and help you set up your operator.