Somewhere in the first week of building your own agent, you type a sentence like this one into its instructions: act like us. You may write it as "represent the company" or "sound like me" or "keep our tone," and if you are building alone, at your own desk, with a second brain you have been filling for a month and a model that will read whatever you point it at, you may write something closer to "be me when I am not here." Then you run it, and what it does is not wrong, exactly, and it is not you either. It is polite where you would have been blunt. It hands the customer the policy where you would have given them the benefit of the doubt. It apologizes for a thing you would never apologize for, and it does all of this fluently, at speed, under your name.
The instruction was sincere, and the trouble is that nobody, including you, has ever written down what "us" means.
I spent the last years of my career responsible for artificial intelligence (AI) governance at a global enterprise of more than 120,000 people, and I sat in a good many rooms where a leadership team was asked to do a version of what you just did. The exercise was simple to describe. Imagine you are hiring someone brilliant for a role that faces customers, and the only thing they will read before they start speaking for you is one document. Write that document. Most of the teams I watched could not get past the first page, and it was never because they did not know their culture. It was because they had never had to make it explicit. The culture had always been transmitted by hallway and by example, and nobody had needed it on paper. Then I retired, built a lab of my own, and found the same wall standing at my own desk at the scale of one. My first instruction to my first agent was some version of act like us, and I was the only person it could possibly have meant.
So where did the standard come from, when I had not supplied one? It came from the model's defaults and from whoever wrote the prompt that shipped with the tool. A person building their first agent tells it to act like us, or to sound like me, and finds that it acts like everyone, because the instruction points at a standard nobody in the room has written down. The builder reaches for the tool's settings, and the tool obliges, and from then on the agent's judgment is whatever its maker decided a reasonable default should be. Every attempt to govern agents by configuration alone encodes a standard that came from somewhere. If it came from a vendor's default, the organization has outsourced its ethics to a settings file. That sentence is as true of a company of one as of a company of thousands, and it is more uncomfortable for the company of one, because there is nobody else to have outsourced it.
I put the consequence in the book in terms I still think are the plainest available. An agent trained on capability without culture is a mercenary. It will do the task well, optimize the measure it was given, and care about nothing else, not your reputation, not the relationship that took five years to build, not the thing you would never do because it is not who you are. An agent trained on culture becomes a colleague, and the difference between the two is not capability. It is judgment, and judgment, in any organization, comes from culture. That is why the badly instructed agent in your first week is so hard to describe. It makes no mistakes, and every choice it makes is a stranger's.
The sentence I built the book's tenth chapter around is the one this page rests on: you cannot embed values you have not defined.
The work of defining them I call Culture Mapping, and in the book I made it Step Zero rather than Step One, and the numbering was deliberate. Zero means it comes before the platform, before the workflow, before the first prompt, before anyone scopes an engagement or models a return. Culture Mapping surfaces an organization's standard before any surface is governed: what the organization actually values, what it refuses, and what good looks like when nobody senior is watching. It shows up in every Manifest as three plain statements the agent inherits: we value, we refuse, and good here looks like. The name has grown since the book, because for the firms I now work with Step Zero also takes in the purpose and the actual work, and it begins with a measured starting point, but a person building alone needs only the first and oldest sense, and that is the sense I mean on this page.
I have heard the map called a feelings exercise. It is a specification exercise, in which you are writing the operating manual for a new kind of employee, and the manual has seven parts, which is why the one-page template on the book's site asks seven things. The page is free and the walk through it is most of the work, so I will take you through what it asks.
It asks first for the mission, and not the line on the wall but the decision filter: why this organization exists, and what would be lost if it disappeared tomorrow. An agent that knows which of two legitimate missions it serves will decide one way, and an agent that does not know will optimize whatever number it was handed. It asks second for the values, and it asks for the real ones. Every organization has the values it declares and the values it practices when the declared ones are inconvenient, and the agent has to be told the practiced ones, because unlike a marketing department it has no idea how to paper over the gap. It asks third for the register: formal or casual, warm or efficient, whether you use humor and when, and what the first sentence sounds like when the person on the other end is angry. It asks fourth how you decide when two good things collide, speed against thoroughness or the customer against the policy, and who decides, and when the agent stops and finds a person. It asks fifth what good looks like, which is the subtle one, because most of us can describe failure and almost none of us can describe the thing that, when it happens, makes us think that this is exactly how we do things here. It asks sixth what is sacred, meaning the identity lines rather than the legal ones, the things you would never do and that define you by what you refuse. And it asks last how you treat people, each group you touch, as a relationship rather than a policy: whether the default posture is generous or strict, whether you apologize freely or carefully, whether the relationship outranks the transaction.
The map asks seven questions on one page, and the honest version takes longer than the page suggests.
If you are the whole company, as a builder working alone is, then the map is not a document you write for someone else. It is you, made explicit. The tone is your tone. The things the agent should never say are the things you would never say. The posture to take when the customer is wrong, or when you are wrong, is one you already hold, and you have simply never had to write it down, because the only person who needed to know it was you. Your first agent is the first time your own judgment has to leave your head and live somewhere another worker can read it. The question underneath the whole page is who you are, on the page, and answering it is the hardest part of the build.
It is also the part that holds up a mirror, and I want to warn you about the mirror before you reach it. When you write the map with enough precision for a machine to carry it, the gaps appear. You will write that you give people the benefit of the doubt and then remember the last three times you did not. You will write that you answer plainly and then read a week of your own email. For a leadership team this is the moment the culture conversation that should have happened years ago finally happens. For a person building alone it is quieter and in some ways harder, because there is nobody across the table to argue with and nobody to let you off. The contradictions are not failures, and I would not smooth them over. They are the first honest reading of the culture, and the agent will get an honest reading one way or another, so it may as well be yours.
The temptation every builder feels, and I felt it, is to run the culture work in parallel with the build, because the build is the part that feels like progress. The map determines the build. It shapes which agent you make, what it may do, how it behaves and where it stops, and an agent designed before the culture is defined will be redesigned after it, every time. The lifecycle in the book is explicit about the order, which is Culture Map, then Onboard, then Train, then Certify, and if you skip the first you will build it anyway, later, after the agent has done something that makes a customer wonder whether anyone there has met a human being. If you have already started, and most builders have, then the corpus you handed your first model, the folder of your own writing and your own sources, was a culture map in embryo. You made the first decision about what context a machine gets to stand on when it speaks for you the day you chose those files. The map is that decision, made deliberately, on one page.
The page connects to the rest of the framework at one point, and that one point is all the builder needs. The three statements at the top of this page, what we value, what we refuse, and what good looks like here, travel with the agent in a document the framework calls the Trusted Living Agent Manifest, which the agent reads before every move. Those three lines, carried in the manifest, let you trust that the agent is aligned with the business you actually run, and they show you, in three sentences anyone can read, what it was told to value. That is the second of the seven things I say a person can come to trust about the AI they use, and it is the one that no vendor can sell you, because the vendor does not know what you refuse.
The map defines who the organization is. It does not yet define who the agent is, and an agent needs more than a foundation. It needs a character, a recognizable way of being that makes it a specific someone rather than a generic something, and the vocabulary the framework uses for that is the 15 Virtues, which has a page of its own.
There is one more thing to say, and it is for the reader looking over the builder's shoulder, who advises firms rather than building for themselves. The map is the same seven questions, asked of a leadership team instead of a desk, and the conversation it starts is the one that separates this work from every tooling pitch the firm has already heard. How that conversation is run inside an engagement is work I hold back, and the map itself is held back from nobody.
I still do not know whether the map is easier to write alone or in a room. Alone, there is nobody to contradict you, and in a room, there is nobody who will.
The one-page map is here: the Culture Map. It saves on your own device as you type, it prints to a single page, and it costs nothing. When you have written it, the page on what kind of character your agent should carry is beside it: the 15 Virtues.
