Stay Confused Without Panicking

I have started to think that one of the most useful skills in modern work is the ability to stay confused without panicking.

This sounds almost embarrassingly basic. Confusion is usually treated as a temporary defect, something that should disappear as quickly as possible. When someone asks us a question, we feel pressure to answer. When a system breaks, we feel pressure to suggest a fix. When a client describes a complicated situation, we begin arranging it into a clean story before we have understood what is actually happening. Silence feels awkward. Uncertainty feels like a public failure of intelligence.

I know this pressure well because I have spent a large part of my life being rewarded for understanding things quickly. In school, speed looked like intelligence. In meetings, the person who produced the first coherent explanation often appeared to be the person who understood the situation best. In design and technology, confidence is constantly mistaken for competence because confidence is visible while careful reasoning mostly happens quietly. You can watch someone present an answer. You cannot easily watch them resist five attractive but incorrect explanations before arriving there.

The problem is that difficult situations rarely become easier just because we are uncomfortable. A broken system does not care that we would like to feel competent again. A confusing medical result, a failed payment, a security incident, or a product behaving differently on two screens will remain complicated while we produce increasingly polished stories about it. Sometimes the stories are plausible enough that we stop investigating. We begin repairing the explanation rather than the problem.

Software debugging makes this especially obvious. When something fails in a familiar way, experience can produce the answer almost instantly. That kind of intuition is real expertise. But when the obvious fixes fail, a different skill becomes necessary. You have to slow down and separate what you observed from what you assumed. You have to ask which parts of your mental model must be wrong. You have to design a test that teaches you something instead of making another random change and hoping the problem disappears.

This is where panic becomes expensive. Panic does not always look dramatic. It can look extremely productive. You open more tabs. You ask three AI models. You rewrite the function. You change the configuration. You restart everything. You generate ten possible causes and try them one after another. There is movement everywhere, but the movement is partly an attempt to escape the feeling of not knowing. Each new action provides a few seconds of relief because it creates the impression that the situation is under control.

AI makes this worse because it can manufacture that feeling almost instantly. It can give us a diagnosis, a strategy, a root-cause analysis, and a patch before we have described the problem properly. It is very good at turning incomplete information into a coherent answer. Coherence feels like understanding, especially when the answer is written confidently and contains the vocabulary we expect. We can move from confusion to implementation without ever spending much time in the uncomfortable middle where we establish what is actually true.

Fast thinking is becoming abundant. The scarce skill is deciding whether any of it deserves to be believed.

There is research behind this distinction, although it does not support the easy slogan that intelligence no longer matters. Cognitive ability helps us process complexity, learn patterns, and solve difficult problems. But researchers also measure partially separate capacities that affect whether we use that ability well.

Cognitive reflection is the ability to notice that an answer arrived too easily and may need to be checked. Metacognition is the ability to monitor whether our own understanding and strategy are working. Actively open-minded thinking includes looking for contrary evidence, considering alternatives, and updating a belief when the evidence changes. Research on the need for cognitive closure describes the pressure to settle uncertainty quickly and then preserve the conclusion we reached.

These qualities are related to intelligence, but they are not the same thing. A person can be quick, articulate, and capable while still accepting the first explanation that restores their sense of control. Another person can appear slower because they are keeping several possibilities alive, gathering information, and refusing to confuse a plausible story with an established fact.

That slower person may look less impressive in the first five minutes. They may be much more useful by the end of the day.

The research also makes the picture less tidy. Slow thinking is not automatically wise. People can deliberate for a long time with bad information and become more confident in the wrong answer. Experts can be extraordinarily fast because they have encountered a stable pattern many times and received enough feedback to learn it. A doctor, engineer, dancer, or designer may see something immediately that a beginner would need hours to understand.

The harder skill is recognizing what kind of situation you are in. Is intuition working because the environment is familiar and your feedback has been reliable? Or are you forcing a strange situation into an old pattern because admitting uncertainty would damage the image you have of yourself?

That image matters more than we admit. Many people have built an identity around being smart. Being smart often means being quick, articulate, and rarely visibly lost. When a problem refuses to yield, it threatens something larger than the task. Taking more time can feel like becoming a different person in front of everyone. So we protect ourselves by answering. We choose the first explanation that makes us feel like ourselves again.

Research on praise for intelligence offers an indirect clue about how this can happen. When success is attached to an identity such as being smart, failure begins to say something about the person rather than only about the current attempt. The evidence comes largely from educational settings, so it should not be stretched into a grand theory of senior engineers. Still, the pattern is familiar. A difficult problem becomes dangerous because it threatens the role we have learned to perform.

I wonder how much better our work would become if we could say, without embarrassment, that we still do not understand what is happening.

There is a version of this sentence that sounds passive: I do not know, therefore I cannot proceed. But there is another version that is active and precise: I do not know yet, here is what I have observed, here are the assumptions I no longer trust, and here is the next test that could reduce the uncertainty. This is not intellectual helplessness. It is refusing to pretend that a guess has become evidence just because the room wants an answer.

The same skill matters beyond engineering. In relationships, panic can make us assign motives before asking what happened. In politics, it turns complicated events into stories with clear villains and heroes before the facts have settled. In medicine, it can lead people either to dismiss symptoms too quickly or to build terrifying explanations from fragments of information. In business, it creates strategies designed to end a discussion rather than describe reality.

Our systems are rarely designed to support uncertainty. Social media rewards the cleanest explanation. Meetings reward the first person who takes a position. Consulting rewards the framework. AI products are mostly designed to produce answers, even when the more useful response would be to expose the missing information. There are few visible rewards for the person who keeps several explanations alive and slowly removes them one by one.

Perhaps this is why clear thinking can feel socially uncomfortable. It requires us to resist the rhythm of the environment. The meeting is moving on. The client wants certainty. The interface asks us to choose. The AI has already produced something polished enough to paste. Remaining uncertain can look like hesitation when it may actually be the only honest response to the available evidence.

Staying confused without panicking is therefore not the same as enjoying confusion. The goal is still to understand. The difference is that we stop treating the emotional discomfort of uncertainty as proof that a conclusion must be reached immediately.

A practical version of the skill is surprisingly simple. State what happened without explaining it. State what you expected. Separate direct observations from assumptions. Keep more than one possible explanation alive. Choose a test whose outcomes would make those explanations easier to distinguish. Predict what the test should show before running it. Then update the mental model rather than only patching the visible symptom.

Every action should either reduce uncertainty, test a belief, or validate the eventual solution. Random experimentation is still useful sometimes, but only when we can say what each result would teach us.

This becomes more important as AI makes output cheaper. We can now generate more possible fixes, arguments, images, strategies, and explanations than we have time to evaluate. The bottleneck moves away from producing an answer and toward maintaining a trustworthy relationship with reality.

The person who stays confused slightly longer may be the person who notices that the two apps are describing the same transfer differently, that the successful test never covered the actual failure, that the elegant strategy depends on a number nobody verified, or that the confident explanation leaves one inconvenient fact untouched.

This is why staying confused without panicking feels almost like a form of resistance. It means accepting that intelligence may sometimes look unimpressive from the outside. It may look like reading the same log again. It may look like asking a basic question in a room full of senior people. It may look like writing known, assumed, and unknown on three pieces of paper. It may look like admitting that the explanation everyone has already agreed on does not account for what happened.

I am trying to become more comfortable in that state. I want to notice the moment when confusion becomes emotional and I begin reaching for any answer that will make it disappear. I want to remain there slightly longer, separate the facts from the story, and let the problem keep some of its shape before I compress it into a solution.

Maybe intelligence is partly the ability to find an answer. But clear thinking may begin earlier, in the moment when no answer has arrived and you manage not to invent one.

Connected reading

• The Real Test of AI Code Is the Next Prompt
• The Prototype Has Become Too Cheap to Trust
• The Most Dangerous Interface Is the One That Says Everything Worked
• The Real Bottleneck Is Human Attention
• The Cheapest Way to Be Wrong

AI authorship note

This essay began with Kris Haamer reading Sean Goedecke's argument about clear thinking and selecting the line "stay confused without panicking" as the Raw Drafts angle. Kris directed the framing, title, tone, paragraph structure, and final publication decision. AI researched peer-reviewed work on cognitive reflection, metacognition, actively open-minded thinking, need for closure, intelligence praise, and expert intuition; drafted and revised the essay; created the SVG illustrations; and handled the publishing workflow.