Journal

Who Signs for It?

Her raise came in lower than she expected. Not insulting, just lower, and lower in a year she believed had gone well. Her manager had told her as much more than once, usually in passing, the way managers do when things are going fine and nobody feels the need to schedule a conversation about it. So she sits down across from him and asks why, carefully, the way people ask when they already suspect the answer will not help.

He turns to his monitor. The number is there with a score beside it, a pay band, and a short paragraph of generated summary about goals and output. He reads a little of it back to her, and then he says the sentence a lot of managers are about to find themselves saying: "The system scored it." She nods, because there is nothing to say back to that. You cannot argue with a score, and you cannot ask a pay band what it was thinking. She thanks him, goes back to her desk, and the conversation is over before it ever really started.

Nothing in that room was a lie. The manager told the truth, and the system did score it. What was missing was anyone who owned the decision. She is not one person. She is a conversation that plays out in companies of every size, and the tool in the middle of it is a lot newer than the conversation itself.

When I write about this topic, AI and the machine are the same thing. It is the word I use when I talk about it, so it is the word I use here. Year-end reviews and raises are a few weeks out. If your numbers pass through a machine before a manager ever sees them, some of your people are going to ask why, and when they do, somebody has to be able to say, "That one was mine." The question is who that somebody is, and the honest way to find out is to walk the raise back up the line and ask everyone who touched it.

The vendor that built the scoring will tell you, and mean it, that their product does not make pay decisions. Their customers do. They sell a tool. HR chose that tool, set the bands, and decided which inputs counted, and HR will tell you, and mean it, that they never decide what any one person gets. They built a process so managers could be consistent. The manager did not build the model, set the bands, or pick the budget. He was handed a number and asked to deliver it. The executive approved the total raise budget for the year and trusted the people below to divide it fairly. That executive has never seen this employee's score and may not know her last name, because deciding individual raises was never that job.

Four people, and four true statements. Every one of them points somewhere else, and every one of them is right about their own piece. A decision about a real person ends up with fingerprints all over it and a signature nowhere. Nobody hid anything and nobody dodged. The work was divided so cleanly that the ownership was divided right along with it, until there was none left for anyone to hold.

I have been on the receiving end of this. In 2003 I was in marketing at Precious Moments, and I was good at it. The e-commerce work I was building had grown past a million dollars a year. Then leadership changed, and between that change and a manager in IT, I was moved out of marketing and into the IT department. My desk went from my own office to a shared space inside a server room, with air conditioners blasting all around me. My work went from building content and the online store to pulling wire and fixing computers that were not really broken. None of it was in my job description, and none of it was my skill set.

Nobody owned it. It was "corporate," in the worst possible way. So I went to the department head myself and said it plainly: I was being asked to do work outside my job, and I was not willing to keep doing it. Move me back to marketing, or let me go. By December I was out the door with unemployment. The only way I got anyone to own that decision was to put my own job on the table. That turned out to be the start of the road that led to the company I sold in 2022. It worked out, but it still does not change how the decision got made.

There was no machine in that story. There was a leadership change and a bad manager. Today there might be a machine, and whoever the decision lands on is in the same spot either way: go find someone who will answer, or stop asking. AI did not invent this. It just gave the oldest dodge in management a machine to point at.

In the first entry of this series, The Decision You Cannot Hand Off, I wrote about the leader alone with a finished answer before it goes out the door. This is the other end of the same decision. It is the moment after, when the person it landed on walks in and asks why and finds that the answer has no address. I named this gap in RemAIning Human. The space between the person who decides and the person who answers for it has always existed in organizations. AI widens it, because it adds a step that looks like a decision-maker but cannot be held to account by anyone. I put it this way in the book: "Authority becomes legitimate when responsibility is visible."

Visible responsibility is easy to agree with in a meeting, and most companies have a policy somewhere that says managers own their decisions. That policy has never once answered an employee's question. A policy cannot sit across the desk from somebody and say, "I looked at this, I made the call, and here is why." Visible means a person, a name, someone sitting in a chair who will say, "That was me," and then stay in the chair while the other person disagrees.

Without that person, a leader still has power. The title still works and the number still goes out. What is gone is the authority, because authority is earned by standing where people can find you when something you decided lands on them. The courts are asking the same question right now, and they have not answered it.

The case most people point to is Mobley v. Workday. Derek Mobley, the lead plaintiff, alleges that Workday's AI screening tools discriminated against him and other applicants on the basis of age, race, and disability. By his account he applied for more than a hundred jobs at companies that used Workday's platform and was turned down every time, often within hours of applying, sometimes in the middle of the night. A rejection that arrives in the middle of the night was not typed by someone at a desk. A decision was made about that man, and by his account, nobody was sitting there when it went out. So far it is only his allegation, not a finding.

Workday disputes the claims. Its position is that its tools do not make hiring decisions and that its customers keep full control of their own hiring. Read that again with the raise conversation in mind, because it is the vendor's line from the chain above, said in a federal courtroom. The court has not settled who is right. It has allowed the claims to move forward on the theory that Workday can be treated as an agent of the employers using it, and it has preliminarily certified a nationwide age discrimination claim to proceed as a collective. A hearing on whether the broader case can go forward as a class is set for March 9, 2027. No court has found Workday liable for anything. The case is open, and anyone who tells you how it ends is guessing.

The government's side is moving too. In 2023 the EEOC published guidance on how existing employment law applies when employers use AI to make decisions about people. In January 2025 the agency took that guidance off its website, and some leaders read that as the rules going away. They did not. Title VII still applies. So do the Age Discrimination in Employment Act and the Americans with Disabilities Act. An employer that uses a tool to help decide who gets hired, promoted, paid, or let go is still answerable under all three, and private plaintiffs can still bring their own suits whether a federal agency is publishing guidance or not.

None of it is settled. The guidance came and went, and the biggest case is still months away from a hearing on whether it can even proceed as a class. Anyone waiting for the law to tell them exactly who answers for an AI-assisted decision is going to wait a long time. You do not need a verdict to answer the question inside your own building, and you can answer it this week: when someone in your company asks why a decision about them came out the way it did, can they find a human being who will answer for it? If the answer is no, you have a problem that no ruling is going to fix for you.

Most companies will never see a courtroom, and the cost shows up somewhere quieter. Think about a company of forty people. There is no legal department, one person handles HR along with three other jobs, and the leader at the top still knows most people by name. Nobody there is going to sue, and nobody is going to file a complaint, which is exactly what makes this dangerous at that size. The damage does not announce itself.

A good employee asks why once. She hears some version of "the system," and she learns that asking does not get her an answer, so she stops asking. She does not quit and she does not complain. Her effort drops a notch, never far enough for anyone to notice and never far enough to get flagged by the same machine that scored her. She still does the job. She just stops bringing the extra she used to bring: the idea in the hallway, the problem she spotted before it became a crisis, the Saturday text that saved a customer.

Her coworkers notice, even if nobody says so. They saw her ask and they saw what she got back, so they learn the same lesson without having to ask for themselves. Questions about decisions go to a machine, and machines do not answer. I wrote in Twenty Points Apart about how small the moment is that starts this. One clipped reaction, and the team files it away.

Before long the information stops moving. The problems that used to come up the line stop coming. They did not go away. The people who would have raised them quietly decided it was not worth the trouble. A leader in that company starts to feel like things are calmer than they used to be, when they are only quieter, and a leader who cannot tell those two apart is in trouble. I keep a rule for this: silence kills more progress than conflict ever could. A forty-person company is where you can watch it happen.

One day she resigns, and nobody saw it coming. In the exit conversation, if there is one, she finally says what she stopped saying a year ago: "I never could get a straight answer about anything." The leader hears that and assumes it must be about something big. It almost never is. It is usually about one small conversation, one raise, one review, where she asked a fair question and nobody would answer it.

That silence is the bill, and it never shows up as a line item. It shows up as turnover nobody can explain, ideas that stopped coming, and a culture where people do what they are told and nothing more. No lawsuit could cost a forty-person company what that costs, and the fix has to come earlier, before anyone sits down at that desk.

Every AI-assisted decision about a person gets one name on it before it goes out: the name of the person who will own the answer when someone asks why. It has to be a person, because a department cannot sit across a desk, HR cannot be argued with, and "the process" has never answered anyone's question. In a raise cycle, that means the manager's name goes on every number on their team, and putting a name on it carries real requirements.

That person has read what went into the decision, meaning the inputs and not only the summary: what the machine weighed, and what it could not have known. They can explain the decision to the person it affects in their own words, and if all they can do is read the monitor back, they are not ready to stand behind it yet. They also have the power to change it. A manager who has to defend a number they are not allowed to change has been made the scapegoat for it. One test comes before the number reaches anyone. Put the employee's question to the manager first: why is the number what it is? If the honest answer is still "the system flagged it," nobody owns that answer yet, and the number does not go out until somebody does.

This is slower. It will add hours to a raise cycle, and it should, because those hours belong to the job you already have. When I wrote about feedback, I described what this part of leadership costs: looking another human being in the eye and saying, "This isn't right." Answering for a decision costs the same thing. You are the one in the chair when the question comes, and you stay there.

The same goes for whoever put the machine in place. Most leaders own something: a team, a process, a department inside a bigger company. If you chose the tool or approved it, you answer for that choice too. That means giving the people under you real power to change a number, and backing them when they do. Otherwise the name on the decision belongs to the person left holding it, not the person who chose the tool. In the first entry, the move was to ask what the output does not know about this person. This one asks something just as plain: whose name is on it?

Go back to that desk. Same employee, same smaller number. This time, when she asks why, the manager does not turn to the monitor. He looks at her and says, "That number is mine. Here is what I weighed. Here is what would move it next year. If I missed something, tell me now." She may still not like the number. She may push back, and he may change it or he may not. Either way, she walked in with a question and walked out knowing who answers for her, and she will ask again next time.

Power has always had somewhere to hide: a title, a policy, a spreadsheet, and now a machine. Each one gives a leader a place to stand just out of reach. Authority does not hide. It stands where people can find it, says "That was mine," and stays in the room while someone pushes back. Owning the answer means agreeing to be argued with, and the pushback is how trust gets built. Nobody trusts a decision they are not allowed to question, and nobody trusts a leader who keeps sending them to a machine.

Your people are going to ask why. Decide now who answers.

When was the last time someone asked you why, and the honest answer pointed somewhere other than you? If you are working out how to handle that in your own company, I would like to hear about it. Let's have a conversation. It's free.