AI Agent Job Description Template

This page gives you the operating document before the service pitch. Copy the blank template, review the completed example, then adapt the fields to one bounded workflow. The goal is not to write a clever prompt. The goal is to define what the agent may do, what it may not do, and who is accountable when something changes.

Every field should narrow authority, not decorate the page
Field	What to write	Control question
Business job	Name one business outcome in plain language.	What useful work is the agent responsible for?
Trigger	State what event or request starts the run.	When should this agent wake up?
Inputs	List the data, records and context required to act.	What must be present before the job starts?
Allowed reads	Name the systems, fields and scope it may inspect.	What may it look at?
Allowed writes	Name the exact draft or write actions it may perform.	What may it change, and where?
Decisions the agent may make	Define the choices it may take without asking first.	Which judgments are automated?
Human review rules	Set the conditions that pause the run for approval.	When must a person approve or reject?
Escalation	State who takes over on missing data, exceptions or risk.	Who owns the handoff when the agent should stop?
Evidence and logs	List the records you must keep for replay and audit.	Can a reviewer reconstruct the action later?
Service target	Set a useful quality, time or throughput standard.	How will you know the job is operating well?
Owner	Name the accountable business or technical owner.	Who accepts the risk and performance?
Review date	Set the next control review date.	When will the authority boundary be checked again?

Copy-ready asset

Use the blank template or start from the completed example

Both blocks below are plain text on purpose. Copy them into your operating docs, ticketing system or agent registry. The blank version is for real use. The completed version shows the level of specificity that makes the template worth using.

Blank template

AI Agent Job Description

Business job:

Trigger:

Inputs:

Allowed reads:

Allowed writes:

Decisions the agent may make:

Human review rules:

Escalation:

Evidence and logs:

Service target:

Owner:

Review date:

Completed example

AI Agent Job Description

Business job:
Prepare a renewal brief for one professional-services account manager before a client renewal call.

Trigger:
Run when an account manager requests a renewal brief for a named client account inside the CRM.

Inputs:
CRM account ID, current renewal date, recent support history, open opportunities, contract record and the account manager's request.

Allowed reads:
Read renewal-stage CRM fields, recent support cases, contract value, current product mix and the last 90 days of account activity.

Allowed writes:
Create one draft renewal brief in the approved workspace and one draft follow-up task in the CRM after human approval.

Decisions the agent may make:
Flag missing data, summarize renewal risks, propose follow-up questions and rank the top three retention issues.

Human review rules:
Pause before any CRM write, client-facing message, renewal-stage change, discount recommendation or task assignment to another owner.

Escalation:
Escalate to the account manager if required fields are missing, account status is unclear, a support case is severe, or a contract record conflicts with CRM data.

Evidence and logs:
Log request time, user, source systems read, records referenced, draft output link, approval decision, final action, errors and recovery path.

Service target:
Produce a usable first draft within 5 minutes with zero unauthorized writes and complete trace logs.

Owner:
Director of Client Success.

Review date:
2026-09-30.

Why define the agent role this explicitly?

Because a job title is not an operating boundary. Adobe's current enterprise agent-components guidance says roles and guardrails define what the agent is for, what systems it may access and when it should defer to a human. NIST's current software-agent identity and authorization work says the risk appears when agents gain access to tools, applications and diverse data. A good job description turns those ideas into one usable record.

If your team cannot fill in the write scope, review rules or owner, the agent is not ready for broader authority. That is not a documentation problem. It is a governance problem.

The point of the template

A strong template does two things at once. It helps the operator launch one useful workflow faster, and it gives reviewers one place to check whether the scope, permissions and approval path still match the business intent.

How should you use the template in practice?

Choose one workflow with a named owner and one clear business outcome.
Fill the template before you widen permissions or enable writes.
Start with read-only or draft-only authority wherever possible.
Bind human review to the sensitive action, not to a generic notification.
Review the record after real usage, then expand only one boundary at a time.

OpenAI's current human-review guidance for the Agents SDK treats approvals as a pause before side effects such as edits, cancellations or sensitive tool actions. Microsoft's current governance guidance says enterprises should standardize approved frameworks and protocols so agent behavior stays secure and predictable. That is why the job description should name the operating boundary before the workflow goes live.

Practical rule: if the template still says "various systems," "standard data," or "as needed," you have not yet defined a safe agent role.

How does this page differ from the governance and approval guides?

AI Agent Governance Implementation is the parent six-control framework. Human-in-the-loop approval that actually controls an AI agent explains how approval pauses must work. This page is narrower: it gives you the one operating artifact you can copy into the workflow record, registry or implementation brief.

Use the template with Custom Agent Deployment when you need help mapping the workflow, protocol boundary and rollout path. If you are still choosing the first use case, start with the AI Workflow Opportunity Mapper.

Which current primary sources shape the template?

Adobe: Core components of an enterprise AI agent, published April 30, 2026.
NIST: New Concept Paper on Identity and Authority of Software Agents, February 5, 2026.
Microsoft Learn: Govern and secure AI agents, last updated June 26, 2026.
OpenAI: Guardrails and human review, accessed August 25, 2026.

Brainiac custom agent deployment

Turn one template into one governed live workflow

Bring the finished job description, the systems involved and the action you want the agent to perform. Brainiac can help define the control boundary, protocol path and rollout sequence for the first live release.

Discuss Custom Agent Deployment

Measured outcome: a qualified conversation tied to one named workflow and one explicit authority boundary.