CREATE YOUR FIRST AGENT BLUEPRINT

Brief your AI like you would brief a new teammate.

Start with the work, define the boundaries, and decide how you will measure value before connecting a tool.

01Job titleThe role, in plain words

02ResponsibilitiesWhat this coworker does every day

03AccessWhat it can see and touch, and what it cannot

04InputsWhat information comes into the workflow

05OutputsWhat it delivers and in what form

06Decisions that stay with youThe judgment a person keeps

07Off-limits data and actionsHard boundaries that cannot be crossed

08Escalation rulesWhen it must stop and hand work back

09Success measuresHow you will know it is helping

Baseline five measures before you build

Write down today’s baseline, then compare it with a small supervised test. The goal is useful improvement, not automation for its own sake.

1 · Time recoveredManual time minus review time

2 · Output qualityEdits or corrections required

3 · Work capturedItems completed or follow-ups found

4 · Error rateMistakes and missed escalations

5 · Cost vs. valueTool cost compared with practical benefit

Your finished agent blueprint

Use this as a planning foundation. It does not configure connections, permissions, or an organization-specific working system.

1. Job title Inbox Triage Assistant 2. Responsibilities Review new messages and identify what is urgent, what needs a response, what is informational, and what may be a newsletter. 3. Access Read-only access to an organization-approved email account. No permission to send, delete, archive, forward, or change account settings. 4. Inputs New email messages, approved priority rules, known key relationships, deadlines, and the organization’s data-handling guidance. 5. Outputs A daily priority brief grouped into: needs attention today, response needed, informational, possible newsletter, and uncertain. Include a one-sentence reason for every priority or uncertainty. 6. Decisions that stay with you Final priority, relationship-sensitive judgment, the meaning or tone of a response, commitments, approvals, and every decision to send, delete, archive, or forward. 7. Off-limits data and actions Beneficiary information, HR matters, health or legal information, donor payment details, credentials, private attachments, and any action that changes the mailbox. 8. Escalation rules Stop and flag messages involving sensitive data, financial or legal commitments, conflict, safety, unclear intent, key relationships, or classification uncertainty. 9. Success measures Staff time recovered; percentage of priority messages caught; number of corrections required; number of appropriate escalations; monthly tool cost compared with time and value recovered.