Guide
Designing AI Roles for Your Team
How to write an AI role the way you would write a job description — scope, inputs, authority, escalation and review.
- Type
- Guide
- Topic
- AI Agents
- Author
- AEGIS OS™
- Published
- Updated
Key takeaways
- A role that cannot be described in one sentence is not a role. It is a collection of unowned tasks.
- Scope is defined by the data and tools a role may touch, not by the instructions it is given.
- Every role needs a named human owner who is accountable for its output.
- Roles are reviewed on a schedule, like people, not left running unexamined.
The five fields every role needs
- Job to be done
- One sentence describing the outcome the role produces, in the language the business already uses.
- Data in scope
- The specific records and fields the role may read, and the ones it may write.
- Tools permitted
- The systems the role may call, and the actions it may take in each of them.
- Approval boundary
- What the role may complete on its own, and what must be reviewed by a named person before it leaves the organisation or changes a committed record.
- Review cadence
- Who examines the role's output, how often, and what triggers an immediate review.
A worked example
| Field | Definition | |
|---|---|---|
| Job to be done | Prepare every new enquiry for a human first review | |
| Data in scope | The enquiry record and published reference material only | |
| Tools permitted | Read the enquiry store; write a draft summary field | |
| Approval boundary | No outbound communication without a named approver | |
| Review cadence | Weekly sample review; immediate review on any escalation |
Written this way, the role can be staffed, audited and retired. Written as a prompt, it can only be guessed at.
Start with the roles nobody wants
- Preparation work: gathering, normalising and summarising before a person looks
- Reconciliation work: comparing two records and surfacing the differences
- Drafting work: producing a first version in the organisation's own format
- Monitoring work: watching for a condition and raising it to the right owner
These four carry the least judgement risk and the most recovered time. Roles that decide, commit or communicate externally come later, and they come with an approval gate attached.
One owner per role
If no person's name sits against a role, its output belongs to nobody — and the first time it is wrong, the organisation has no defined path to correct it.
Sources and references
- AEGIS AI workforce enablement — Published description of how roles are scoped and enabled.
- Architecture Assessment — Where roles, owners and boundaries are documented for one organisation.
Reviewed and published. No client example, vendor comparison, pricing claim or performance figure is described.
Start with an Architecture Assessment
An Architecture Assessment turns these questions into a documented blueprint, readiness analysis, implementation roadmap and quote for your organization.
Related resources
Framework
Operating Boundaries for AI Agents
The six-part specification every intelligence role carries before it is allowed to touch live work.
Framework
AI Agents vs Automations vs Workflows
How agents, automations and workflows differ in scope, control and accountability, and when each one is the right instrument.
Framework
What Should Never Be Fully Automated
Decision classes that require a named human owner, and how to encode those boundaries in an operating system.