Using AI to Reduce Friction

"Clear thinking becomes clear writing; one can't exist without the other." — William Zinsser

A friend tells you at dinner that AI is overrated. You can tell they mean it, because they spent two hours that afternoon proving it to themselves. They watched a tutorial, opened a chat window, typed what they wanted, got something that ran, hit an error, asked for a fix, hit another, asked again, hit a third, and ended further from the thing they wanted than when they started. The conclusion wrote itself. The tool does not work.

The tool worked fine. It produced exactly what they asked for at every step. What it could not do was figure out what they actually wanted, because they had not figured that out either, and no amount of fluent output supplies a clarity the asker never brought. The machine is an amplifier. Feed it clarity and it multiplies clarity. Feed it confusion and it multiplies confusion, faster and more convincingly than you could have managed on your own.

It builds fast and cannot decide

What AI does astonishingly well is collapse the distance between having an idea and having a working version of it. That distance used to be enormous. The path from idea to prototype was so long and so expensive that most ideas died untested, because nobody could afford to find out whether they were any good. Now the loop is an afternoon, and that genuinely changes what a person can attempt in a year.

What it cannot do is decide what the thing should be, or whether it is good, or whether the working version fits the situation it has to live in. AI is so fluent it feels like it is doing the thinking too. It is not. The bottleneck was never the coding or the writing or whichever skill the tool just made cheap. It was always clarity, and making the cheap skill free simply left clarity standing in the open, with nowhere left to hide.

The skill it left standing

So the skill that suddenly matters more than any of the ones AI absorbed is the oldest one there is. Saying what you want clearly enough that any competent helper, human or machine, could act on it.

A prompt is just a description handed to something with no other context, and that last part is what people forget. The model does not know what you have tried, what you already know, what you are working under, or what you meant by the word you left vague. Whatever you leave out, it fills with a guess, and the guesses are the bad output you blamed on the tool.

The fix is structure. Tell it the situation, in a couple of honest sentences. Tell it the goal, as something more than make this better. Tell it the format you want back, and the constraints it has to respect. Sixty seconds of that routinely buys hours on the other end.

It works the same across a conversation, where the most useful thing you can say is, before you answer, here is what you should know about my situation.

The same law runs the exam room. The patient who walks in able to say I have been off for three weeks, worse in the afternoons, worse after meals, here is what I have tried gets a useful visit. The patient who can only manage I just feel bad spends the whole appointment in twenty questions before anything can start. Same doctor, same minutes, different outcome, decided almost entirely by how clearly the situation was described. The model is just a faster, more literal version of that same mirror.

The three ways people waste it

Almost every disappointing session is one of three failures of that clarity.

The first is handing the machine the thinking. Write me a marketing plan asks the model to make the entire decision, and it cannot make a decision it has no information to make. Break it into the smaller asks where you stay the one deciding, and it turns useful immediately.

The second is giving it no context, the help me with this code with no code and no goal and no error attached, which forces a guess and then gets blamed for guessing.

The third is taking the first answer. The first output is the model's best guess for the average person in the average situation, and the real answer almost always lives in the second or third turn, after you have pushed back on what the first one got wrong. The people who close the tab after one mediocre reply never find out what the thing can actually do.

Underneath all three is a single rule that keeps the amplifier pointed the right way. Do not accept output you cannot explain. If you cannot say what it does or why it solved the problem, you do not understand it yet, and understanding is the one thing that cannot be outsourced no matter how good the tool gets.

Put this to work. Before the next afternoon like your friend's, spend the first sixty seconds writing down the situation, the goal, the format and the constraints, then push back at least once before you take an answer.

The tool that applies this. The AI Workflow Map sorts the tasks where AI earns its place from the ones where the clarity is still yours to bring. → Member Library