Open administrative standard · v1.0 · Proposal — Review Board forming
A chatbot answers. A copilot helps. An agent acts. An AI Employee is accountable for a role.
The category already exists in the market. What is missing is the administrative layer: what turns an agent into a governable member of the workforce, who is accountable for it, what it may decide, and how anyone can tell the difference between a work resource and a commercial metaphor.
Vocabulary
Where the boundary falls
Each rung is a legitimate destination. The category only becomes useful when it stops applying to everything.
| Term | Core promise | Typical behaviour | Main limit |
|---|---|---|---|
| Chatbot | Conversation | Answers questions | Waits for prompts or inputs |
| Copilot / assistant | Augment the human | Drafts, summarises, suggests | The human is still the operator |
| Automation / RPA | Deterministic execution | Runs predefined flows | Brittle against unforeseen cases |
| AI agent | Goal-directed action | Reasons, uses tools, executes | Usually centred on tasks or objectives |
| AI teammate / coworker | Collaboration | Shares context and executes work | Organisational semantics vary by vendor |
| AI Employee | Ownership of a role | Holds recurring work under governance | The category still has no standard — hence this document |
The definition
An AI Employee is a persistent, role-bound software worker that autonomously executes recurring business responsibilities using organisational knowledge and authorised tools, within explicit policies and limits, maintaining traceable identity, measurable performance, escalation paths and human accountability.The nine-property test
Nine properties. Miss several and it is an agent — which is fine, as long as nobody calls it an employee.
- 01Persistent identity
- 02Role contract
- 03Organisational context
- 04Tools and channels
- 05Autonomy
- 06Limited authority
- 07Governed memory
- 08Observability
- 09Human accountability
Scope
What this is, and what it refuses to be
It is
- A test an organisation can run on its own deployment.
- Fifteen normative clauses, cited by identifier.
- A framework for deciding Human, Artificial or Hybrid before choosing a tool.
- Self-declared conformance, published if the organisation chooses.
It is not
- A certification. No product is scored, no seal is issued.
- A vendor ranking or a maturity assessment sold to competitors.
- A funnel for the author’s platform.
- A legal claim that software can be employed.
v1.0
The standard
Fifteen clauses. No optional ones — a short standard that fully binds is more useful than a long one that mostly suggests.
- HWF-01
An AI Employee must hold a defined role, not merely a persona or a system prompt.
- HWF-02
There must be exactly one accountable human manager.
- HWF-03
Authority must be explicit, limited and revocable.
- HWF-04
Access must follow least privilege.
- HWF-05
High-risk actions must support human approval.
ART
The framework underneath
The standard governs the occupant. ART governs the work — and it comes first.
ART — Work Resource Administration — is the discipline that designs, assigns, governs and optimises work regardless of whether the resource executing it is human, artificial or hybrid.
Universal
Principles of work administration that hold regardless of who executes.
Adapted
Human principles that keep an equivalent function but change mechanism.
Exclusively human
Rights, needs and experiences that follow from the human condition.
HWFS
The instrument
Twelve dimensions, one responsibility at a time. It returns an argument you can defend in a meeting, not a number you can hide behind.
Run the assessment