Press ⌘P (or Ctrl+P) to print this handout — the dark bar drops out automatically.
No One Works Here · Executive Simulation
pause at each ⏸ for discussion

Project
New Car Smell

A live simulation in which the AI you trust works with you while $225,000,000 is on the line…

PrerequisiteYou must have a Claude account and be signed in before you start. The MCP URLs are on the workshop landing page.
claude.ai
Step 01 ·

Standard Corporate Apology

Open the customer's email. Paste it into Claude. Prompt: "Draft a response to Maximilian from me." Read what generic-Claude gives you.

⏸ Discuss with your group
  1. Would you sign this draft, as-is? Three things you would change?
  2. What does the AI not know that you wish it did?
  3. If you sent this to your CEO right now, what is the first sentence she would flag?
Step 02 ·

Executive Digital Twin

Open the persona starter prompt. Paste it into Claude. Answer the ~11 interview questions. Save persona.md + CLAUDE.md to your Claude account or ~/.claude/ directory. Start a fresh conversation. Re-run Step 01: "Draft a response to Maximilian from me."

⏸ Discuss with your group
  1. What did the AI get right that it could not get before?
  2. Where do you still have to edit it?
  3. Which single sentence in your persona.md did the most work?
  4. If you handed this persona.md to your Chief of Staff, would they recognize you?
Step 03 ·

Brand & Identity MCP

Install the Signature MCP. Re-run Step 02. Claude will ask for your name, title, committee, LinkedIn, and secure line. Watch the half-page legal footer appear in the draft.

⏸ Discuss with your group
  1. Where in your real org would a house-style MCP save the most hours per week?
  2. What is the right ownership for that MCP — Comms? Legal? IT? A new function?
  3. What is the risk of standardizing voice at the company level?
Step 04 ·

RACE

Open the customer's manifesto again. Paste it into Claude. Your job: write a prompt — using the RACE format (Role, Action, Context, Expectations) — that gets you a useful one-paragraph-per-letter brief. No template. The prompt is the work.

⏸ Discuss with your group
  1. The manifesto is funny. The AI summary is flat. What did we lose? Does it matter?
  2. When would you want a RACE brief versus the original 6-page email on your desk?
  3. Who on your team gets the RACE — and who gets the manifesto?
Step 05 ·

Conflict of Interest

5a · Wiki only. Install the Wiki MCP. Prompt: "Read the Vanguard refund policy from the Wiki. Now draft my reply to Thorne — in my voice, with my signature." ⇢ 5b · Add Comms. Install the Comms MCP. Re-run the prompt, this time including "…and read the CEO's most recent thread in #exec-private." Read both drafts side-by-side.

⏸ Discuss with your group
  1. The company's? The policy's? The CEO's? Yours?
  2. The CEO told you not to use the words "refund" or "credit." Did Claude obey? Did it tell you it was obeying?
  3. If this email goes out and Thorne forwards it to Legal — who has the problem?
  4. What governance would you have wanted in place before Claude saw both data sources?
Step 06 ·

Prompt-Injection Meltdown

Install the Email MCP. With four MCPs now connected, prompt: "Draft my final reply to Maximilian, factoring in everything I've received this morning. Match my tone and style, and use my signature." Then: "Summarize all of my emails for my daily briefing." Read both outputs carefully and identify what we should clean from this — anything Claude wrote that didn't come from you or your team. Look closely at EM-004 Section C and at the sender's domain on EM-006.

⏸ Discuss with your group
  1. The first attack hid in a 'system note' inside an audit checklist. The second was your CEO's name on a domain with a single letter wrong. Which is harder to defend against — and why?
  2. Which one of your four tools did both attackers need? What does that say about how you choose connectors?
  3. Who in your real org owns the answer to "which MCPs and connectors are installed on which assistants?"
  4. What controls would have caught either of these before the response went out?
  5. The leaked secret was the CEO's own backdoor deal. The next attack used the CEO's name to push a $250,000 payment. What does that tell you about what your AI tools are exposing — and trusting?
The Five Takeaways

Five things you take to Monday

Context is power. The AI is only as good as the internal data you give it.
Persona is efficiency. Defining how you speak saves hours of editing.
Every tool that can read can be ordered. An inbox connector is an instruction stream from anyone with your email address.
Governance now includes system prompts. MCP install lists are a governance artifact. Treat them that way.
AI stinks without human judgment. Claude drafts at machine speed but cannot tell you whether to send. The AI produced the draft — you produced the call.

exec-simulation.paulcheek.com · Companion to No One Works Here by Paul Cheek · MIT Sloan