Lesson 10 of 12 · Change

Human integration and change management

Agentic AI changes job design, not just tooling. If you don't specify how humans supervise agents, especially in act-with-approval mode, your organization will either distrust the system or use it recklessly.

New roles and workflows

Reviewer and approver responsibilities and SLAs.
Exception-handling playbooks.
"When to trust, when to verify" guidance.
Escalation protocols when confidence is low.

Specific training

How to write instructions that agents execute reliably.
How to interpret citations and tool traces.
How to supervise actions, especially in act-with-approval mode.

Update the SOPs

If the agent's workflow and the humans' workflow diverge, you get inconsistent outcomes and accountability gaps.

Even great agents fail if the organization doesn't adapt how work is done.

Put it to work in step 10 Lesson 9 Lesson 11

Check your understanding

3 questions. Pick an answer for each, then check. Nothing leaves your browser session.

1. What happens if approvals are slow?

Nothing; the agent waits

The time saved by the agent is lost, and incidents can take longer than before

The agent approves itself

2. What two failure modes does the article predict without supervision rules?

Distrust or reckless use

High cost or low speed

Too many or too few agents

3. Why update SOPs alongside the agent?

To avoid accountability gaps when agent and human workflows diverge

For the auditors only

Because models read SOPs

Answer every question to check.