How it works
How to put mandates to work
An AI agent can do useful work without unlimited permission. It starts with a person granting specific authority, and ends with a named owner for every action that matters.
Think of a mandate as the working agreement between a person and an AI agent. The person defines the job, the limits, and the decisions that must come back to them. The agent contributes within those boundaries. RealMandates is building software to make that agreement part of the workflow, and to keep a clear record of who owns each consequential action.
- 1
Set the mandate
A named person decides what work an agent is allowed to do, how it shares the work with people, and when it must come back to a person. The mandate carries that person's name.
- 2
Start with the agents you have
Begin with one agent your team already uses and one kind of work it does.
- 3
Name the decisions that matter
Agree which actions are consequential: the ones that reach customers, spend money, change records, or speak for the company.
- 4
Bring in the right person
A budget change, a customer offer, or a pricing exception reaches the accountable owner with the context needed to decide.
- 5
Keep the record
Every consequential action keeps the signal that started it, the mandate it followed, the decision made, and the person who owns it.
Illustrative example
From buying signal to customer outreach
A target account starts researching a new capability. An agent drafts a personal message to the buyer. Its mandate lets it prepare drafts but not send outreach to strategic accounts without review. The account owner edits one line and approves the send.
- Signal
- Target account researching a new capability
- Mandate
- Draft only for strategic account outreach
- Decision
- Approved with one edit
- Owner
- Account owner
What a mandate should answer
- Who granted the authority?
- What is the agent allowed to do?
- Which actions need a person first?
- Who is accountable for the outcome?
One question to bring to a design partner conversation
Pick one real agent workflow in your business. Name the next consequential action it could take. Who can authorize it today? What would the record need to show tomorrow? That is where a useful conversation begins.
Discuss your use caseWhy is this more than an approval workflow?
Because the mandate begins before the approval request. It describes the agent's authority, the kinds of work it can perform, the boundary for human judgment, and the accountable person. Approval is one possible step in a broader governed workflow.