Every genuinely hard call I've made in this business was easy in the abstract and hard at four o'clock on a Thursday with a client waiting.
That gap is the whole problem. In the abstract you're reasoning. In the moment you're tired, behind, and looking for a reason that the thing you already want to do is fine.
You will find one. People always do. The mind is very good at producing arguments for conclusions it has already reached, and it is fastest when you're under pressure.
The only defense I've found is to decide beforehand, when nothing is at stake and you have no incentive to be clever.
Why written matters
"I know what I believe" is not sufficient. I've known what I believed and then discovered, at four on a Thursday, that what I believed was more flexible than advertised.
Writing it down does something specific. It creates a version of you, with no deadline pressure, who gets a vote later. When the moment arrives, the argument is no longer between you and an abstraction. It is between you now and you then, and you then was thinking more clearly.
It also makes drift visible. If you break a written rule you have to notice you're breaking it. Unwritten values bend silently and you never catch the moment they moved.
Ours
Here is the framework we use at Connexion AI. It is short on purpose. A values document nobody can recite is decoration.
People matter most. Humans take priority in every AI decision. When a choice trades human good for efficiency, the human wins. This is the one that costs money, which is how you know it is doing work.
Values guide decisions, not the reverse. You don't start from what the tool can do and reason toward what is acceptable. You start from what you believe and evaluate the tool against it. Capability is an input, never a justification.
Responsibility requires accountability. No scapegoating the tool. "The AI wrote it" isn't a defense and never was. If it goes out under our name, it's ours.
AI should promote human flourishing. Not output. Not efficiency. Flourishing, meaning people are actually better off in ways they would recognize. Those come apart more often than you would think.
I'm not offering these as the right four. They are ours, arrived at by argument, and yours should differ because your work differs. Steal the format, not the content.
The exercise
Take twenty minutes. Not a retreat, not a committee.
Write three to five sentences about how you will and won't use AI. Fewer than five. If you can't remember them, they won't help you in the moment, which is the only place they matter.
Make at least one of them expensive. A values list that never costs you anything is a marketing document. Ask yourself which one of these would I break first if the money were good enough, and then write that one down more carefully.
Be specific enough to violate. "We use AI ethically" can't be broken because it doesn't mean anything. "We never send client-identifying data to a tool that retains it" can be broken, which means it can also be kept.
Say where the hard line is. For most people there's some category of work they will not take. Naming it in advance is much easier than discovering it mid-negotiation.
Then tell someone. Your team, your business partner, your spouse. A commitment nobody knows about is one you can quietly revise.
What this is for
I want to name the thing underneath, because otherwise this reads as corporate hygiene.
The reason to decide in advance is not risk management. It is that the person you become is made out of small decisions taken under mild pressure, most of which you won't remember. Nobody drifts because of one big compromise. They drift because a hundred four-o'clock Thursdays went the easy way and each one seemed reasonable.
Writing your values down is a way of making a promise to yourself before you have a reason to break it. It doesn't make you good. It just means that when you do the wrong thing, you have to do it on purpose, with your eyes open, having read the sentence you wrote.
That turns out to be a fairly high bar. Most of the time it's enough.
Your challenge, and I gave this to a room of students this summer with a Monday deadline: write yours this week. Three to five sentences. Before you need them.