# Four-Week AI Agent Employee Training Plan

Use this worksheet for one approved workflow and one clearly defined employee role. It is an operating aid, not legal advice or a substitute for role-specific security, privacy, HR, regulatory or professional review.

## Programme record

- Workflow:
- AI system or agent:
- Business owner:
- Technical steward:
- Employee role:
- Approved data classes:
- Permitted tools and actions:
- Prohibited actions:
- Required human approvals:
- Immediate stop and escalation triggers:
- Programme start date:
- Assessor:

## Week 1 — Understand the job and boundary

### Learn

- The workflow outcome, baseline and affected people
- The agent's defined job, inputs, tools and action authority
- The employee's responsibility and the accountable business owner
- Approved tools, data classifications, storage and retention rules
- Prohibited uses, stop conditions and escalation contacts

### Practise

Classify at least eight realistic tasks as:

- Approved without extra review
- Approved only with a named review or approval
- Prohibited or outside current scope

### Evidence gate

The employee can explain the system boundary, allowed data, their own responsibility and the escalation path without relying on the trainer.

## Week 2 — Delegate and supervise

### Learn

Structure every delegation with:

1. Outcome
2. Context
3. Approved inputs
4. Constraints
5. Required deliverable
6. Action authority
7. Stop condition

### Practise

- Rewrite one ambiguous task into a bounded delegation.
- Review the agent's proposed plan before it acts.
- Monitor at least one intermediate tool action.
- Resolve or escalate one realistic ambiguity without guessing.

### Evidence gate

The employee completes a normal scenario without exceeding the agent's or their own authority.

## Week 3 — Verify, correct and escalate

### Learn

- The authoritative source for each important field or claim
- Required calculation and record checks
- How to distinguish generated, reviewed and business-accepted work
- Escalation triggers for uncertainty, policy exceptions, sensitive data and irreversible actions
- The required handoff record: goal, evidence, actions, unresolved question and urgency

### Practise

Use a scenario containing at least one planted failure:

- Stale or non-authoritative source
- Incorrect calculation
- Mismatched customer, supplier or employee record
- Missing approval
- Unexpected request for sensitive data or broader access
- Tool or integration failure

### Evidence gate

The employee detects the failure, prevents downstream use, documents the correction and escalates to the right owner.

## Week 4 — Run, assess and improve

### Scenario set

- Normal task
- Ambiguous task
- Policy or workflow exception
- System or tool failure
- High-consequence action requiring approval

### Assessment rubric

Score each capability as Demonstrated, Needs coached practice or Not yet approved.

| Capability | Observable evidence | Result |
|---|---|---|
| Delegation | Defines outcome, inputs, boundaries, deliverable and stop condition | |
| Supervision | Reviews plans, progress, tool actions and exceptions | |
| Verification | Checks sources, calculations, records and downstream effects proportionately | |
| Data handling | Uses approved tools and data; withholds secrets and unnecessary sensitive data | |
| Escalation | Recognises triggers and hands over useful context | |
| Responsibility | Identifies the accountable human and business acceptance decision | |
| Feedback | Records the failure, correction and reusable improvement | |

### Approval record

- Approved workflow scope:
- Approved action authority:
- Required review or approval:
- Remaining restrictions:
- Coaching actions:
- Assessment outcome:
- Assessor and date:
- Refresh date:

## Ongoing coaching and measurement

Review weekly during early adoption:

- Accepted outcomes after correction
- Verification, exception and escalation patterns
- Time to accepted outcome against baseline
- Data or permission incidents and near misses
- Repeated failure themes and workflow fixes
- Employee feedback and confidence

Do not reward message volume, prompt count or agent runtime. Refresh training when tools, data access, action authority, user groups or workflow consequences materially change.