The Leadership Friction  |  Edition No. 28  |  September 8, 2026

Your Team Needs an AI Operating Agreement

AI policy tells people what they can’t do. It rarely tells them how the team is actually supposed to work together, and that gap is where the real risk lives.

Many organizations are talking about AI policy right now. That conversation matters. But policy alone doesn’t answer the question leaders actually need answered.

A policy tells people what tools are approved, what information can and cannot go into a system, where the legal and compliance lines sit. These are important and necessary questions.

They just aren’t the only ones. The one underneath them all is simpler: how are we actually going to use AI together? That’s what a policy can’t give you. That’s what an operating agreement is for.

A Policy Sets Boundaries. An Operating Agreement Sets Expectations.

A policy tells people what they can’t do. An operating agreement tells them how AI fits into the way the team actually works.

When is AI appropriate to use, and when does a decision need a human? What gets reviewed before it leaves the building? When should someone say out loud that AI helped create the work? What kinds of calls should never get handed off to a tool, and who’s responsible for checking the output when it’s wrong?

Those sound like small operational questions. They aren’t. They quickly become leadership questions, because without a shared answer, everyone on the team starts writing their own rule book.

People Are Already Writing Their Own Rule Book

This is the risk I see most often with AI adoption. Employees don’t wait for the organization to sort it all out. They experiment. They find tools they like, build their own prompts, and decide what’s safe to type in and what needs a second look.

I understand why. But when everyone is operating on their own logic, the organization doesn’t actually have a standard. It has as many standards as it has people.

One person reviews every AI-generated document line by line. Another assumes that if it reads well, it’s accurate. One manager uses AI to prepare for a hard conversation. Another lets it write the whole conversation and follows the script word for word. One employee knows exactly what counts as confidential. Another is guessing.

The technology is consistent. The behavior around it is not.

Where Leadership Has to Step In

You don’t need to control every AI interaction your team has. You do need to stop leaving your people to guess.

Start with what AI is actually for in your organization. Research, brainstorming, drafting, meeting summaries, customer communication, the answer may shift by role, but nobody should be inventing their own answer because you never gave them one.

Then there’s the question of what gets checked before it goes out the door. AI helping someone brainstorm is not the same as AI producing something that touches an employee, a customer, or a financial decision. The bigger the consequence, the stronger the review needs to be, and that standard has to come from you, not from whoever happens to be doing the work that day.

Confidentiality is where I see the biggest risk. Customer information, employee data, financials, strategy; don’t assume your people already know where that line sits. Be clear because silence gets read as permission.

There’s also a real difference between AI making a recommendation and AI making a decision. Those are not the same thing, and the distance between them matters most in hiring, performance management, pricing, and anything touching a client relationship or your broader strategy.

And when AI helps produce the work, the person submitting it still owns the result. That doesn’t change. What changes is how the work got done.

AI adoption is moving faster than organizational clarity. In one 2024 survey, 56% of organizations had no policy governing generative AI use at work. Meanwhile, Microsoft’s 2024 Work Trend Index found that three out of four people were already using AI at work, often with tools they brought in themselves.

The Goal Isn’t More Rules

An AI operating agreement shouldn’t become another document nobody opens. It should clarify the work, and it can do that in a page or two.

We use AI to assist thinking, not replace judgment. The person submitting the work owns its accuracy. Sensitive or confidential information stays out of tools we haven’t approved. High-impact decisions require human review. When you’re not sure, ask before you use the tool.

Five sentences can prevent a lot of confusion.

What Happens When All They Hear Is No

Your team is watching how you introduce AI. Some people are excited about it. Some are worried about what it means for their job. Some genuinely don’t know what’s expected of them. And some are already using it and hoping nobody asks.

If leadership only shows up with rules and restrictions, that last group grows. People stop disclosing and start hiding, and a hidden AI problem is a much harder one to manage than an open one.

The fix isn’t more enforcement. It’s an honest conversation.

• Where do you want people using this?
• Where does it help? What skills matter more now?
• What still has to stay in human hands?

That conversation does as much work as the agreement itself.

Instead of asking your leadership team “do we have an AI policy,” ask this instead: if I asked five people on our team how they’re expected to use AI in their work, would I get five similar answers?

If the answer is no, that’s an expectations problem, and it gets more expensive to fix the longer it sits.

If I asked five people on your team that question right now, what do you think they’d say? I’d genuinely like to know what you find.