AI GOVERNANCE SIMULATOR
Run your workflow before anyone builds it.
Somebody on your team is about to automate something with AI, and nobody can say what it will read, what it will send, or who signs off. Name the tasks that eat your week, in your own tools, and this page plays the first one step by step: the trigger, what the AI reads, what it drafts, who approves, what goes out. Flip the approval off, or put the AI on a personal account, and watch what changes. Every task you name gets a verdict (automate, assist, or keep with a person), the steps in the tools you use, the line the AI must never cross, and the hours back from your own numbers. It works with Microsoft 365, Google Workspace, QuickBooks, ChatGPT, Gemini, Claude, and whatever you type in. No email is needed.
SEVEN QUESTIONS
Name the tasks. The page does the rest.
Pick the team, tick your tools, tick the tasks that eat the week (as many as six, or type one of your own), and say whether customer data is involved. Run it once per team if you are looking at several. The simulator and the blueprint are written on this page the moment you finish, and they get an address of their own. Tool and task names only in the text boxes: no customer records, no credentials.
Or open a worked example first: a sales team of 2 to 5 on Microsoft 365 and HubSpot, a finance team of 6 to 15 on Google Workspace, QuickBooks and ChatGPT or one person doing HR on Microsoft 365 with no AI tool yet. Each opens with its answers filled in and its result written, so you can change one answer and run it again.
Keep a copy in your inbox.
Enter a work email, company and role to receive this blueprint as text with its address. This records an SIT inquiry for follow-up, and the engineer reads it before any call. Marketing emails require a separate choice.
YOUR REGISTER
What this browser has worked out so far.
Kept in this browser only; nothing on this list is stored on our server. Each line opens its own address, so you can come back to it, send it to the owner, or clear it. The procurement pack binds every line into one PDF, rebuilt from the addresses when you press the button, with a cover and a contents list.
WHAT IT IS
What should AI governance look like?
Five steps you can point at: a trigger, what the AI reads, what it drafts, a named person who approves, and what goes out, with the log of who asked and who signed.
Governance is not a document. It is where the AI stops and a person starts, written down before anyone builds. The simulator on this page shows those five steps for your own task in your own tools, and shows what happens when one of them is missing.
THE INSIGHT
Why run it before anyone builds it?
Because the mistake that costs the most is made at the read step, before a single message goes out: an AI on a personal account that reads a customer record has already carried that record outside your boundary.
Run it first and the boundary is chosen on purpose. Build it first and the boundary is whatever the first person who tried it happened to use.
HOW IT WORKS
How does the simulator write the result?
From a rules table, not a model: each task you tick has a pattern (draft, summarize or decide), a data class and a place where it acts, and those three set the verdict, the steps and the line the AI must never cross.
- You pick your function, your tools, the tasks that eat the week, and whether customer data is involved.
- The page plays the first task in five steps and lets you flip the approval, the AI account, the data marking and the log on and off.
- Every task gets a verdict, the steps in your tools, the never line and the hours back from your own minutes.
- The result gets its own address. Nothing is stored unless you ask for the email.
ANY STACK
What if we use Google Workspace, QuickBooks or ChatGPT?
Then the steps are written for those tools, and the safe half names the training and retention settings in the AI tool you use, because the five steps are the same in every stack and only the settings differ.
Microsoft 365 Copilot returns only content the signed-in user already has at least view permission to see, so anything shared too widely is findable from day one. (Microsoft Learn, checked Oct 31, 2026)
Microsoft 365 shops get the Copilot boundary and the permission review. Google shops get Drive sharing and Gemini activity. A mixed shop gets both, and anything you type that the page does not know gets its own row in the foundation, where the settings are checked by hand.
YOUR DATA
What happens to the answers?
They are rendered on the page and encoded in the address, and nothing is stored unless you ask for the email, in which case the lead is recorded through the same intake as the contact form.
Do not enter customer records, credentials or anything you would not put in an email. The free-text box is for tool names only.
THE MONEY
What is free and what is paid?
The blueprint is free: the verdicts, the steps, the never lines and the hours. Making it true in your tools is a ten-hour engagement, and the build is scoped from there.
Ten hours buys read-only discovery, the signed permission row, the gates switched on, and the five controls marked tested instead of assumed. No price appears on this site; the engagement is quoted after a scoping call.
AFTER TWO WEEKS
What will be working after two weeks?
- You have watched your own first task run with a person at the gate, and watched it run without one.
- You have watched the hours move under the form as you changed a minute, a frequency or a head count, before the page wrote a step, and you can open a worked example and change one answer instead of starting blank.
- You have watched the two incidents that happen in practice, oversharing and an agent that did what inbound content told it, in your own tools, each with the control that stops it and the ten questions the owner answers in the first hour.
- Every task you named carries a verdict, the steps in your tools and the line the AI never crosses.
- The hours back are written from your minutes, your frequency and your headcount, so you can defend the number.
- The result has its own address, and the engineer reads it before any call.
HONEST ABOUT LIMITS
What this does not do yet
It does not know your tools from the inside. Copilot Cowork carries out multi-step tasks across Microsoft 365 and pauses for the user's approval before sensitive actions such as sending an email or posting in Teams. (Microsoft Learn, checked Oct 31, 2026) That pause is the approval step the simulator plays. Whether the equivalent exists in your plan and is switched on is the first thing the ten-hour foundation checks.
It does not test anything in your tools: a setting is named, not checked, until read-only discovery checks it.
It does not build the workflow or quote a price.
THE ONE THING TO REMEMBER
Run the workflow before anyone builds it, and the boundary is a choice instead of an accident.
Said another way: five steps, one named person, one log, in your tools.
Said a third way: the blueprint is free; making it true is ten hours.
DIRECT ANSWERS
Questions before you decide.
Do I need an email to see the result?
No. The result renders on the page and has its own address. The email is only for keeping a copy and it records a lead through the normal intake.
Is the result written by an AI?
No. It comes from a rules table, so every verdict and step can be traced to the task and the tools you picked. A model may tailor the wording in a later version, under the same rules.
Can I run it for more than six tasks?
Six at a time. Each run gets its own address, so run it again for the next six. Six is already more than most teams can change in a quarter.
What is the difference from the Governed AI Blueprint?
The blueprint is about one workflow and your five controls. The simulator is about which tasks to hand to AI first and what each one needs. The result links to a blueprint written for its first task.
How is the work priced after this?
Quoted after a scoping call. The foundation is ten hours at every size of firm and the build is scoped from it. No price appears on this site.
NEXT STEP
Start with a boundary you can explain.
Choose a focused check, or speak with the team about the workflow you need to improve.