Take it back to your office

The 30-second refresher

The five levels

1 Chatbot · 2 Assistant with tools · 3 Workflow · 4 Agent · 5 Multi-agent

For any given task, which is the lowest level that would do the job?

Where is the human

In the loop On the loop Out of the loop

Where do the stakes call for a person to stay in or on the loop? For rights or safety, the Office of Management and Budget (OMB) requires human oversight, intervention, and accountability: in or on the loop, not out.

The agent loop

Reason → Act → Observe → Reflect

An agent repeats this loop toward a goal. Whether it also has a stopping condition and a human checkpoint is a design choice, and a policy question.

Your takeaways

☑

Check your AI policy

A quick self-check: does your policy cover what an AI is allowed to do on its own? See your gaps and next steps.

Open the self-check → ►

Keep learning

Short, vetted explainer videos, curated readings, and every source the session cited, right below on this page.

Watch, read, and check the sources ↓ ▤

Revisit the walkthrough

Step a request through all five levels again, and see where the human stays at each one.

Open the walkthrough →

Keep learning

Videos to watch, readings and tools to explore, and the sources the session cites.

Possible questions to take to your team

For a given task, what is the lowest level that would do the job?
When a vendor's material says "agentic" or "AI-powered," what does the tool actually do on its own? Where would it sit on the five levels?
Where should the human stay? Is the AI referencing your organization's own data, and do you know which sources, specifically?
Do you have an AI use policy, and does it cover agentic AI, the levels where the AI can act on its own?
For your vendors: where does their tool sit on the spectrum, who is accountable, and does your policy cover third-party AI?

Your one next step

Two minutes, in your own words. Why write it down? In a meta-analysis of 642 tests, forming a specific if-then plan (naming when, where, and how you will act) improved follow-through across outcomes.

Name one small, low-risk thing you could pilot, and one safeguard you would want in place first (your data, a human check, a log). And one step for your team, if any: what would you share, and with whom?

Copied

Share it

QR code for ai-agentic.pages.dev/takeaway

This page: ai-agentic.pages.dev/takeaway. The QR and the short link also sit on the closing slide, so everyone can open this page on their own device. A short follow-up email a week or two later will point back here.