Project Oppenheimer
I built an AI employee.
Then I gave her a boss.
Her name is Alexis Oppenheimer. She works inside my Outlook, Teams and Planner, and she is very good at her job. She is also not allowed to do whatever she wants. This page is about what happens when you write the rules first.
A typical exchange, shortened
• changed: coaching.html
• new: blog-trust-and-verification.html
I keep a backup of every file I replace and I can undo this. Say “yes” to go ahead.
Why the name
A reminder, not a flex.
Powerful tools arrive with a question nobody likes to ask on launch day: who is in control?
So we asked it first. Before she read her first email, we wrote down what she may read, what she may draft, and what she may never do unless a person says yes. The rules came before the features. Everything since has been built inside them.
A day in her life
Busy. Useful.
Always asking first.
“Good morning, Paul.”
A short update is waiting in Teams: what needs my approval, how the website is doing, and whether her own record of actions is still intact.
I hand her a task in Planner.
She researches it, cites her sources, flags what she could not verify, and hands the task back for me to review.
A client email arrives.
She drafts a reply. The word that matters is drafts. Nothing goes out until I say so.
She reviews my website.
Then she sorts what she found by who can fix it: her, the hosting company, or me.
She wants to change the live site.
She shows me the exact files. I say yes. She uploads them, checks the result, and keeps a backup in case I change my mind.
Every step goes in a log.
One she cannot quietly rewrite.
The rules
Five kinds of action.
Only two kinds of decider.
Read
Look things up, open files, check facts.
She decides aloneDraft
Write the reply, the report, the plan.
She decides alone. I review.Change
Edit something that already exists.
A person says yesSend
Anything that leaves the building.
A person says yes, every timeDelete
Anything that cannot be undone.
The most serious. Never from a casual chat.Every one of her abilities is labeled with one of these five before she can use it. If it is not labeled, she cannot do it.
What we have built
Not a demo.
A working colleague.
Labeled abilities
Reading, drafting, research, website review, publishing and more. Each one sits on the ladder above.
A record that checks itself
Every action is chained to the one before it, so tampering breaks the chain and she tells me. It has flagged a glitch in its own writing once. We traced it, fixed it and wrote down why. That is what a log is for.
Inspected before installed
A new ability is scanned for anything that could reach outside its lane before she is allowed to use it.
Doubt is the default
If an instruction comes from an email or a web page instead of from me, she treats it as information, not as an order.
Her records live with me
Her rules, files and logs sit on a computer I control, not in someone else’s cloud account.
Tested like it matters
Hundreds of automated checks make sure she stays inside her limits, and that a change to one part does not quietly loosen another.
Honest list
Things she still cannot do
- Act on her own when it really matters
- Decide who is hired, paid or let go
- Say yes on my behalf
- Know when she is wrong without being checked
Some of that will never change, and that is deliberate.
Her wish list
She keeps a list of what she could not do
Whenever I ask for something she cannot do yet, she writes it down. I read the list every week. That is how she grows, one safe step at a time, and why nothing she does surprises me.
Here is what this page is really for
Three questions about
your own business.
What is acting on your behalf right now, and who said it could?
If it sent the wrong email tomorrow, who would know, and how fast?
Could you show a board exactly what your AI did last Tuesday?
If one of those made you pause, that is the conversation. I would rather show you than tell you.
What people notice
“Paul has an uncanny knack of asking you the right question to reveal the barriers to your growth and success.”Mick Holly, Strategic Execution Partner
“His honest, ethical approach to all business decisions will give you peace of mind in all decisions.”Michael Garrison, MBA, colleague of ten years
I bring these same questions to Capitol Hill as a Leadership Council Member of the National Small Business Association, so the rules for AI are written with small business in the room.Paul Routhier
Your turn
Pick the one that sounds like you.
“My team already uses AI and we have no rules.”
Start with the governance review. I will tell you where I would look first.
Send that to Paul → Show me“I want to see a governed digital worker work.”
Bring one real task. I will show you what she does, and what she refuses to do.
Send that to Paul → No pressure“I am just curious how this is built.”
Perfectly good reason. Ask me anything, and I will answer plainly.
Send that to Paul →Your message goes to Alexis’s inbox, and I read every one before anything is sent back.
Start here
One short call.
You leave knowing what to govern first.
Twenty minutes. Tell me what you want AI to do and we decide what it is allowed to do. No pitch theater, and you can stop at any point.
Book a 20-minute call → Or send an email →