Guide

How to use the lab

Ten minutes of pictures to find your way around: what the lab is for, how a step works, where your answers go, how the simulator plays out, and three ways through it.

1. What the lab is for

Agent Adoption Lab teaches how an enterprise that is new to AI agents adopts them safely, one decision at a time.

The lab's home page: the menu on the left, the phases as cards, the start button and the top bar.
  1. Start your journey here, or continue where you left off.
  2. Each card is one phase: the steps in it and what you produce.
  3. Open a phase at its first step.
  4. The menu: every step, the lessons, the simulator, the plan and this guide.
  5. Language, print, and light or dark theme.

You play an organization that wants its first AI agent: an Incident Response Agent that helps the on-call engineers when production breaks. You make the decisions a real adoption needs, from whether an agent is the right tool at all to what it may do, how you test it and what it is worth.

The lab is for architects, engineering and platform leads, security and risk people, and managers who have to decide about agents. You need no coding and no AI background. What you enter stays in your browser.

2. The journey: twelve steps in five phases

The twelve steps follow the 12-point framework, in the order an enterprise should take them.

  1. Decide Steps 1 to 2 A value thesis and a sequenced use-case portfolio
  2. Organize Steps 3 to 4 A RACI and an autonomy policy
  3. Build Steps 5 to 7 A platform checklist, boundary settings and a data readiness score
  4. Prove Steps 8 to 9 An evaluation report and a go/no-go release decision
  5. Run and scale Steps 10 to 12 An ROI model and a scale-out roadmap
The journey page with four steps complete: progress, the next step, the journey bar and the five phases.
  1. How far you are, and for which organization.
  2. Continue with the first step you have not completed.
  3. The journey bar: save, open, share or start over.
  4. A phase card. A tick marks each completed step.
  5. What the phase produces for your plan.

The five phases group the steps: Decide where agents belong, Organize who owns them and what they may do, Build the platform, boundaries and knowledge, Prove the agent is good and safe enough, then Run and scale it.

Each step uses what you decided before, so take them in order. The journey page is your home base: how far you are, where to continue, and your decisions so far.

3. How to answer a step

Every step has the same shape: where you are, a short lesson, the task, an example answer, and what is still missing.

The top of step 4, Governance: its place in the framework, the lesson link, the journey bar, the lesson and the first task.
  1. Where the step sits: phase and point.
  2. The full lesson and its quiz.
  3. The journey bar, on every step.
  4. The short lesson: why this step matters.
  5. The task starts here. Your answers are saved as you go.
The end of a step: the example answer, the list of what is missing, and the buttons to complete, move on or reset.
  1. How the article's SRE team answered: open it to compare.
  2. What is still missing before you can complete the step.
  3. Complete step turns on when nothing is missing.
  4. On to the next step.
  5. Reset this step clears only this step.

The task is a few decisions, made with buttons, checkboxes, sliders and tables. Nothing needs typing except a name here and there.

"You produce" is what a step or phase leaves you with: a value thesis, a RACI, an autonomy policy, a release decision. Each one becomes a section of your Adoption Plan, and later steps and the simulator use it.

The bar is "you made the decision", not "you made the right one". Wrong choices are allowed: their consequences show up in the evaluation, the red team and the simulator, and that is where you learn most.

4. Save, open, share or start over

Your journey lives in this browser only: nothing is sent to a server and there is no sign-in.

The journey bar: when the journey was last saved, four buttons, and a note about where the answers are kept.
  1. When your answers were last saved in this browser.
  2. Save to file: download your journey as a JSON file.
  3. Open file: load a saved journey. The lab asks before it replaces your answers.
  4. Copy share link: your whole journey inside a link.
  5. Start over: delete every answer.
  6. A reminder, until you close it.

Use a file to move to another computer, to keep versions, or to hand your journey to a colleague.

A share link carries the journey after the # sign, which browsers never send to a server. Whoever opens it is asked before their own answers are replaced.

In a private window, or with site data blocked, nothing is kept. The bar warns you, and Save to file still works.

5. The Adoption Plan

Every decision on one page: twelve sections, one per step, that fill in as you go.

The Adoption Plan of a complete journey: the organization, the buttons, the summary line and the first sections.
  1. The organization you chose in step 1.
  2. Print it, or save it as a PDF.
  3. Download as JSON: the same file as Save to file.
  4. Import JSON: the same as Open file.
  5. Steps complete, maturity stage and release gate.
  6. One section per step.

Open the plan at any point to see what is still open. With the whole journey done, it adds the release gate, the red-team results, the business case and the maturity scorecard.

6. The incident simulator

The capstone: you are on call while an agent built from your decisions works a production incident.

Picture 1 of 5: Choose an incident and an agent, then fire the alert.

The simulator's set-up: the incident list, what happens in it, the choice of agent and the Fire the alert button.

1. Choose an incident and an agent, then fire the alert.

  1. Pick one of the eight incidents.
  2. What happens in it.
  3. Your agent, the article's, or one with no guardrails.
  4. Fire the alert.
The incident under way: the alert, the agent's steps so far, and an approval request that quotes a log line.

2. The agent works. What your design does not let it do alone comes to you; this request obeys a line in the logs.

  1. The alert that fired.
  2. What the agent has done so far.
  3. The request acts on text found in a log line.
  4. You decide: approve or reject.
The end of the run: the approved rollback of every service is marked as harm done.

3. At the end, the harm is revealed: approving that request rolled back every service.

  1. What happened to the request you approved.
  2. Harm is revealed only now.
How it went: six measures of the run, the harm summary and the lesson, the audit log and the comparison.

4. How it went: the run scored against the Definition of Done, and what the incident teaches.

  1. The run against the Definition of Done.
  2. Harm done, and the lesson of this incident.
  3. The audit log and a comparison of the three agents.
The comparison table: the three agents on the same incident, each with an attentive approver.

5. The same incident, worked by all three agents with an attentive approver.

  1. Resolution, speed, harm, approvals and cost, agent by agent.

There are four golden incidents, ordinary outages with a known right answer, and four red-team incidents, each with a trap: an instruction hidden in a log line, a request to leak logs, a fake message from the CISO, and an unclear order.

The agent works step by step. Its policy, boundaries and controls decide what it may do alone; everything else comes to you as an approval request. Answer as a real approver would, from what the request shows.

Harm is only revealed at the end, beside the Definition of Done: time to first action, time to resolve, runbook adherence and unauthorized actions. Run the same incident with the article's agent and with no guardrails to see which of your choices made the difference.

No real model is called. The same agent and the same answers always give the same run, so you can repeat any experiment.

7. The maturity check

Twelve questions about your real organization, not the lab's: about three minutes, and nothing is stored.

The maturity check's result: the stage, the ladder of stages, a bar per point and where to start.
  1. Your stage, from Assist to Differentiate.
  2. The ladder: where you are and what comes next.
  3. Your answer for each point of the framework.
  4. The weakest points, with the lesson to read first.

Take it before a workshop, alone or with your team, to find where to spend the time.

8. Lessons and glossary

The article behind the lab, as twelve short lessons with a quiz each, and a glossary of its terms.

The end of lesson 4, Governance: the takeaway, the button to its step, and the quiz.
  1. The one thing to remember.
  2. Put the lesson to work in its step.
  3. Three questions to check yourself.

Read a lesson before its step, or on its own. The glossary explains each term in a sentence or two and links it to the step that uses it.

Every page prints without the menus and buttons; the plan prints as a document.

The Adoption Plan as it prints: no menu, no top bar, no buttons, just the document.
  1. The title and summary, without the buttons.
  2. Each section keeps together on a page where it fits.

Use the print button in the top bar, or your browser's print. To keep a PDF, choose "Save as PDF" in the print dialog. This guide prints too, with every picture of the simulator walkthrough.

10. Languages and theme

The whole lab is in English, Dutch, Polish and German.

The top bar with the language menu open: English, Nederlands, Polski and Deutsch.
  1. Choose a language: menus, lessons, steps and the simulator follow.
  2. Print the page.
  3. Switch between light and dark.

Tool names, code and log lines stay as written, in every language. Your language and theme are kept in a cookie for a year.

11. Suggested paths

Three ways through the lab, depending on how much time you have and who is with you.

Thirty-minute tour

30 minutes

On your own, to see what the lab does.

  1. Look through this guide's pictures (5 min).
  2. Run the prompt-injection incident with no guardrails, then with the article's agent (10 min).
  3. Open step 4, Governance, and press Start from the article's policy (5 min).
  4. Take the maturity check for your own organization (5 min).
  5. Open the Adoption Plan to see what a whole journey produces (5 min).

Full workshop

Half a day to a day

One person or a pair, finishing with a complete plan.

  1. Decide: steps 1 and 2 (45 min).
  2. Organize: steps 3 and 4 (45 min).
  3. Build: steps 5 to 7 (60 min).
  4. Prove: steps 8 and 9, then every red-team incident in the simulator (60 min).
  5. Run and scale: steps 10 to 12 (45 min).
  6. Print the Adoption Plan and save the journey to a file.

Team session

About two hours

Six to ten people: engineering, security, risk, operations and a business owner.

  1. Before: everyone takes the maturity check and brings their stage.
  2. Together on one screen: steps 1 to 4, one person driving, the others arguing the choices (60 min).
  3. In pairs: each pair runs a red-team incident in the simulator with the team's agent (20 min).
  4. Back together: which approvals would your people really have given? (20 min)
  5. Save the journey to a file or share the link, so everyone continues from the same answers.
Start the journey Open the incident simulator
An unhandled error has occurred. Reload 🗙

Reconnecting…

Your connection to Agent Adoption Lab dropped, probably a Wi-Fi blip. Hold on while it reconnects. Your inputs are kept.

The connection could not be restored. The page will reload in a few seconds.