Why Every Company Needs Guiding Principles for AI

AI is moving faster than any set of rules can keep up with, and that calls for a different kind of leadership. It’s no longer just about communicating what we allow and what we don't. We also have to consider how we want people to think when the rules run out.

In the last article in my AI Leadership series, I wrote about turning what your teams learn about AI into company-wide agreements. Agreements are the foundation. They tell your people which AI systems can work with company information, who can see what, where human review happens, and who's accountable.

But every agreement has an edge. It covers the situations you foresaw. The rest comes down to judgment.

With AI, “the rest” is getting bigger every month. Your people are making decisions about tools, information, and customer-facing work that didn’t exist as decisions a year ago. No leadership team can write agreements fast enough to keep up. Try, and you’ll end up with a rulebook so long nobody reads it and so out of date that the people closest to the work learn to ignore it.

Guiding principles give a company direction beyond the edge of its agreements. And I believe AI makes them more important than they’ve ever been.

What Is a Principle? 

A principle doesn’t tell you what to do. It tells you how to decide: what to weigh, what to prefer, and what’s off the table. That lets a capable person facing a situation nobody anticipated reach the decision the company would have wanted without ever having to ask.

Principles often get confused with Core Values. Both are short lists, and both end up on an office wall. But they do different work: Core Values tell your team how to behave. Guiding Principles tell them how to judge.

You read a value by watching how someone shows up. You read a principle by watching how someone decides, especially when the decision is hard.

Why AI Raises the Stakes

Companies coordinate in three layers. Early on, your relationships do most of the work. You’re in every room, and people know what you want because they heard you want it. As your company grows, agreements take over: Seats, commitments, processes, the things you can write down. And as pace and ambiguity rise, principles become the layer that carries the company through everything agreements can’t reach.

Which layer a company needs depends on the environment it’s operating in. As pressure, complexity, and pace increase, the right layer of coordination has to evolve with them.

AI raises pace and ambiguity at the same time. Tools change monthly. Capabilities show up that weren’t possible last quarter. And the decisions are being made at the edge: by the person in Customer Success deciding whether to trust a summary, by the analyst deciding whether a new tool is worth trying, by the manager deciding how far to let an agent run. Those people are usually two or three layers below anyone who could write an agreement for them.

Without principles, they have two options: They can ask you, which makes you the bottleneck, or they can guess, and everyone guesses differently.

Agreements First, Then Principles

A lot of leaders make the same mistake here. They read about judgment over rules, like how it sounds, and tell the company to “Use good judgment with AI.” In a company with no approved tools, no clear review points, and no owner for its agents, that sentence means “Guess what I would have done.” And that’s exactly how people hear it. They're left to interpret the leader instead of following a clear system.

A principle is a way to decide at the edge of an agreement. But there has to be an agreement for there to be an edge.

That’s why agreements come first. They define the boundaries everyone needs to understand. Guiding Principles come next. They help people make good decisions when they reach situations those agreements don’t fully cover.

Here’s a simple way to tell which is which: If you’d be upset to learn that two people handled the same situation differently, it needs an agreement. If you’d expect good people to handle it differently depending on the context, it needs a principle.

3 Tests for a Working Principle

Too many principles sound good on paper but fall apart when someone has to make a decision. “Be responsible with AI” is a good example. No one disagrees with it, which is exactly the problem.

A working principle passes three tests:

• It rules something out: A principle has to narrow the choices in front of you. If every reasonable person would make the same call anyway, you probably have a platitude, not a principle.

• It can lose: A real principle reflects a preference between two good options, and following it should sometimes mean giving something else up. If it always supports the choice you already wanted to make, it hasn’t really been tested.

• It can be said in the moment: A principle that takes a paragraph to explain can’t be used at the point of decision. It needs to be short enough to say out loud in a meeting as the reason: “We’re not rolling that tool out to everyone yet. Fewer and finer.” That’s the whole mechanism. A principle is doing its job when someone can invoke it in a sentence and everyone knows what it rules out.

Keep the list short. Five to seven is plenty. More than seven turns into a policy manual.

Find the Principles You Already Hold

You don’t have to invent your principles. Most of them are already sitting inside decisions you’ve made and phrases you already use.

Think about the AI calls you and your leadership team have made over the last few months. The tool you approved and the one you didn’t. The report you let AI draft on its own, and the customer email where you insisted a person review every word. The experiment you shut down and the one you celebrated. Then:

1. List the hard calls.

2. Group the ones that point to the same idea.

3. Identify the rule that would have helped people make those same decisions without you in the room.

4. Test it against the next situation that comes up.

If the rule actually decides the question, you have a principle. If everyone can still argue their side under it, keep digging.

Two warnings. First, don’t copy another company’s principles. They came from someone else’s decisions, and the first test will show your people that you don’t decide that way. And second, don’t have too many. Keep the few that settle the questions that keep coming back.

How Our Principles Apply to AI

At Ninety, we have five Guiding Principles. None of them were written for AI, and that’s exactly the point. Principles written for today’s AI would be out of date by next quarter. Good principles can hold up when the situation changes. Here are ours:

• Fewer + Finer: Progress comes from reduction before expansion. It’s what keeps us from rolling out every new tool at once, and it’s why we start AI in one part of the business before we spread it to the rest.

• Truth > Comfort: It’s comfortable to assume an AI output is right, and even more comfortable to assume someone else already checked it. We name what we see instead. When an agent produces something surprising, we don’t explain it away. We find out why.

• Judgment > Rules: We would rather be surrounded by people capable of strong judgment than people who need a policy for every decision. That’s why our AI agreements are short. The bright lines are firm. Everything else trusts the person closest to the work.

• Return > Cost: Every action should produce at least one clear return, and you should be able to say which. That’s why, earlier in this series, I said I’d want to see a cost-benefit ratio of at least 1:5 before standing up an agent. Agents add complexity. The return has to be worth it.

• Succeed ∨ Escalate: There’s no third option. If an AI output looks wrong or someone runs into a situation our agreements don’t cover, they either fix it or raise it clearly and promptly with someone who can. We own reversible decisions and escalate irreversible ones. Noticing a problem and walking past it isn’t neutral. It’s abdication.

That last one does more work than any AI policy we could write because it answers the question every agreement eventually leaves open: What do I do now?

Give Your People a Way to Decide

AI will keep creating situations no agreement could have anticipated. Bright lines give people the boundaries they need, while Guiding Principles help them make sound decisions once they move beyond those edges.

That combination is what lets you push decisions toward the people closest to the work and still operate as one company. Agreements give you reliability. Principles give you judgment that scales.

In the next article in this series, I’ll look at another important question: How do we know whether AI is actually improving the way we work? We’ll look at how to measure its impact in ways that connect back to the outcomes your company already cares about.

AI will keep creating situations none of us planned for. The companies that handle them well won’t be the ones with the longest rulebooks. They’ll be the ones whose people know how to decide.