Professionalized AI Employee

PAE · Category Definition · Evaluation Framework · Evidence of Trust

AI + Human = the smallest employee unit. The human is the subject of responsibility, AI is the component of capability — capability is governed by the employment system, responsibility returns to the human.

Definition in one sentence

A Professionalized AI Employee — AI + Human forming the enterprise's smallest employee unit: under the support of the HR system, responsibility has an owner, capability is compounded, and performance is amplified.

Why now: the wave of failure

Over 40% of agentic AI projects will be canceled by the end of 2027 (Gartner)
The primary cause is not technology but governance — rising costs, unclear value, insufficient risk control
Only about 21% of enterprises have a mature agent-governance model

The root cause is a misalignment: capability unprecedentedly strong, yet without a subject of responsibility. Procure AI as a tool, and the tool acts autonomously; expect AI to bear responsibility, and it cannot become a party to an employment relationship. The answer lies between — attach AI's capability to an employee.

The Six Elements of professionalization (entry threshold)

All six must be present. Missing any one means it is not this category at all:

Element 01

Organizational identity

On the roster: which employee, what AI capability, in which jurisdiction
When absent: Uncontrolled proliferation, no attribution

Element 02

Role authorization

Headcount: what role, what authority boundary
When absent: Overreach or over-restriction

Element 03

Pre-employment certification

Qualified by assessment before taking the role; certificate has an expiry
When absent: Problems surface only after production incidents

Element 04

Tenure audit

Every decision leaves verifiable evidence
When absent: Incidents cannot be traced or attributed

Element 05

Promotion and demotion

Trust level adjusts dynamically with tenure evidence
When absent: Binary governance of full lock or full trust

Element 06

Retirement and inheritance

Experience sediments into organizational assets on departure
When absent: All-or-nothing shutdown, asset loss

The Three Pillars (evaluation framework)

On what basis does an enterprise hand a formal role to an AI Employee? Three inseparable dimensions:

P1 · Trustworthy

No overreach, no loss of control, and every step independently verifiable

Question answered: Can we entrust it?

P2 · Competent

Can truly complete the role, not merely intercept

Question answered: Can it do the job?

P3 · Controllable

However strong, final decision authority stays with the human

Question answered: Who decides?

The Seven-Step Method

A task, from a human instruction to a provable result, follows a structured execution program — plan first, then execute, verify while executing, and leave evidence throughout:

Understand intent

instruction translated to structured task in milliseconds

Formulate a plan

high-risk steps identified up front

Assemble basis

knowledge, experience, and conduct injected at once

Reason and decide

reasoning visible in real time

Rule gate

every action adjudicated: allow / correct / escalate

Execute

controlled execution, results feed the iteration

Audit and anchor

verdict and result become evidence, independently recomputable

The professionalized lifecycle

Seven stages, mirroring a formal employee, each with institutional semantics and an auditable mechanism:

Identity

organizational record: a unique identity credential

Role

occupation blueprint: assemble capability by role

Training

sandbox assessment: certified to work

Operations

the Seven-Step Method: perform the duties

Audit

tenure evidence: anchored into the audit chain

Trust

evidence-driven rating: authorization scales dynamically

Retirement

audit sealed: experience distilled back

Relationship to ERDL: three orthogonal layers

ERDL is the language of the governance layer, MCP is the protocol of the connection layer, A2A is the protocol of the communication layer — three different layers, orthogonally complementary, not competitive:

Connection layer

MCP

Can the agent reach tools and data?

Communication layer

A2A

Can agents collaborate with each other?

Governance layer

This category + ERDL

Can the organization hire, audit, and hold this AI Employee accountable?

Verification: neutrality is measured, not claimed

Why trust it? 318 independently recomputable vectors across two layers:

Decision-evidence layer: cryptographic verification (tamper-proof, recompute JCS + SHA-256)

240

Expression-kernel layer: semantic verification (correctness, evaluate against the spec)

independent runners (Go / Python), rebuilt from the public spec alone, byte-for-byte identical

Neutrality is not claimed — it is independently measured. You can fool people; you cannot fool math.

Full whitepaper

The complete whitepaper (category definition · evaluation framework · evidence of trust · roadmap · glossary): English (on-site) · 站内全文（中文）