Subject: It Was Clear in My Head Artificial intelligence is becoming more capable. That does not mean it is becoming better at reading my mind. Sometimes I give an AI a task and receive something polished, complete, and wrong. It may answer a different question than the one I meant to ask. It may optimize for speed when I wanted care, rewrite something I wanted reviewed, or produce a technically correct answer that is useless for the decision I need to make. The frustrating part is that it may have followed my instructions. I simply did not give it the instructions I thought I had. As AI becomes capable of doing more things, the number of plausible directions it can take also grows. A vague request does not leave it with no path. It leaves it with many paths and asks it to guess which one I had in mind. When the guess is wrong, I can blame the tool. Sometimes the tool deserves the blame. But sometimes the problem began before it responded. The goal was clear to me, so I mistook that private clarity for communication. I learned the same lesson repeatedly over the course of my career. I could see the result I wanted. I knew why it mattered, which details were important, which methods had already failed, and what a successful outcome would look like. All of that information was freely available to me because it was already inside my head. Then I would ask another person to “take care of it,” “improve it,” or “make sure it works.” When their interpretation differed from mine, I could feel as though they had missed something obvious. From their perspective, they were working from the information I actually gave them. I was judging their result against information I had never made available. Communication cannot transfer an intention directly from one mind to another. I have to give the other person material from which they can reconstruct it. The more that remains hidden, the more they have to supply from their own experience, priorities, and assumptions. Competence does not remove this problem. It can make the problem harder to notice. The more capable I believe someone is, the more tempted I am to think they should know what I mean. I provide less context because I trust them to fill in the gaps. They often can fill them in. They just may not fill them with the same answers I would. Clear communication is not necessarily long communication. A page of explanation can still avoid naming the action I want. A short request can be clear if it gives the other person what they need to understand the result, their role, and the conditions that matter. Before I make a request that matters, I want to answer three questions. What is my goal? This is the change I am trying to create. It explains why the request exists and helps the other person make useful decisions when my instructions cannot anticipate everything. What exactly am I asking this person to do? A desire is not yet a request. “I need support” leaves the form of support hidden. “Would you read these two pages and tell me where you became confused?” gives another person an action they can understand, accept, refuse, or discuss. How do I want them to approach it? This includes the constraints that matter, the methods I want used or avoided, and the decisions they are free to make. If I care about the process but mention only the outcome, I should not be surprised when someone reaches it in a way I dislike. I should specify the path only when the path affects the result. If I am asking someone for their expertise, prescribing every step may prevent them from contributing it. Clarity does not require me to control the process. It requires me to distinguish the fixed conditions from the choices I am leaving to them. A useful request can often begin like this: I am trying to accomplish... I would like you to... Please approach it by... For example: I am trying to help the team decide whether to approve the pilot. I would like you to reduce this presentation to ten slides. Keep the cost and safety evidence, organize it around the decision, and make the remaining design choices yourself. That structure is simple. The harder work happens before I speak. I need to interrogate my own request from outside the mind that created it. If someone followed my words literally, could they produce a reasonable result that would still disappoint me? What information feels obvious only because I already know it? Which parts are requirements, and which are merely preferences? What would I need to see in order to say, “Yes, this is what I meant”? If the answer exposes something important, it belongs in the request. For complicated work, one final step helps: ask the other person to tell me what they understood. This is not a test of their intelligence. It makes their reconstruction visible while correction is still cheap. With an AI, I can ask it to restate the goal and plan before beginning. With a coworker, I can ask what they see as the next action and what the finished work should look like. Clarity does not guarantee agreement, cooperation, or success. A person can understand my request and refuse it. They may see a better method, lack the capacity, or disagree with the goal. A clear request leaves room for that response. If refusal is not allowed, I am not making a request. I am giving a demand a friendlier name. Clarity also does not remove responsibility from the listener. They can pay attention, ask questions, and say when something remains uncertain. Communication is work on both sides. But my intended meaning begins with me. No listener, coworker, partner, friend, or machine can extract an instruction I have left inside myself. When a result misses what I wanted, I can begin with a useful question before deciding who failed: Did I clearly communicate the goal, the action, and the path? If I did, then I have something concrete to examine together. If I did not, the frustration may still teach me what the next request needs to contain. It may have been clear in my head. The work of communication is making it available somewhere else.