Being Excited About Your Own Idea Is a Cognitive Impairment

You have an idea. It seems good. You are excited about it.

This is the worst possible state in which to evaluate it.

Enthusiasm suppresses the machinery you would need to assess your own thinking. You notice the reasons it will work and skim the reasons it won't. Ambiguous evidence gets read as support. The obvious objection occurs to you, gets a quick answer, and never comes back. A founder who has spent months on a product feature will, for instance, interpret early user tests that show mixed results as "mostly positive" because the positive signals align with the effort already spent. The negative signals receive less scrutiny because they threaten the narrative already forming in their head.

None of this is a character flaw. It is documented and it applies to careful people. Confirmation bias means you seek and weight evidence for what you already believe. Sunk cost means the more you have invested in an idea, the harder it becomes to abandon. And motivated reasoning means that when you want a conclusion to be true, you will find a route to it and the route will feel like analysis. These patterns show up in controlled studies across domains: investors overweighting data that supports their portfolio decisions, scientists giving more lenient methodological reviews to papers whose conclusions match their prior hypotheses, and managers rating employee performance more favorably when they personally championed the hire.

The traditional fix is other people — a colleague who will tell you the truth. The problem is that such colleagues are rare, busy, and frequently invested in something themselves. Even when they exist, the social cost of delivering unwelcome feedback often leads them to soften the message or wait until the moment has passed.

The Actual Value Proposition

Most writing about AI at work is about output: drafts, summaries, emails, code. That is real, and it is not the interesting use.

The more valuable application is thinking, and the reason has nothing to do with intelligence. It is that the thing on the other side of the conversation is not emotionally invested in your idea.

It will not spare your feelings because you seemed excited. It has no stake in the project. It will not go quiet because disagreeing with you is socially expensive, and it will not agree because it wants the meeting to end. In most human conversations, the listener is simultaneously tracking the content and managing the relationship. A direct report weighing whether to challenge their manager's plan must calculate the risk of appearing difficult. A peer asked for input on a project they might later be asked to support must consider how their comments will be remembered in future meetings. These calculations happen automatically and shape what gets said.

That is a genuinely scarce property. Most of the honest feedback available to a person is filtered through a relationship, and the filtering is invisible to both parties.

What This Looks Like In Practice

Four uses, in rough order of value.

Stress-testing. Not "what do you think of this" — that invites agreeableness. Instead: argue against this as strongly as you can. What is the strongest case that this fails? What would have to be true for this to be a mistake? The framing matters enormously. Asked to evaluate, you get a balanced-sounding assessment. Asked to attack, you get the objections you were avoiding. The difference appears quickly in practice. A product manager who asks for general thoughts on a new pricing model typically receives a few supportive comments and one or two mild concerns. When the same manager instead requests the strongest case against the model, the objections become specific: the assumptions about customer segments that would need to hold, the competitor responses that would make the change unprofitable, and the internal coordination costs that were previously treated as minor.

Thinking out loud. Explaining a problem in enough detail to be understood is most of solving it. This is the rubber-duck effect, except the duck asks follow-up questions and notices when your explanation contains a gap. The value is produced by your own articulation, not by the responses. Many people discover that the act of describing the constraint they had been treating as obvious forces them to examine whether it is actually fixed or negotiable.

Decomposition. Large problems paralyse because they are undifferentiated. Breaking one into parts that can be worked separately is a specific skill, and it is mechanical enough to be assisted while remaining substantive enough to matter. The assistance works best when the user still owns the judgment about which parts are worth separating and which remain entangled.

Rehearsing difficult conversations. The performance review, the disagreement with your manager, the client call you have been putting off. Working through what you want to say, and what the other person is likely to say back, before you are in the room. This is one of the highest-value uses and almost nobody does it. The rehearsal surface issues that only appear once both sides of the exchange are simulated rather than imagined in the abstract.

The Ways This Goes Wrong

The failures are all human failures, and they are predictable.

Treating the output as an answer. The most dangerous mistake, because the tone is uniformly confident whether or not the content is correct. It does not say "I'm unsure" or "I may be inventing this." Confidence and accuracy are simply not correlated, and the fluency is what makes it hard to remember that. In a thinking context this matters differently than in a drafting context: a wrong fact in a draft gets caught in review, while a wrong premise in your reasoning propagates into a decision. Once a premise is accepted, subsequent steps are built on top of it without re-examination.

Outsourcing the judgment. There is a real difference between help me think about this and tell me what to do. The second feels efficient and hollows out the thing you were supposed to be developing. If you cannot reconstruct the reasoning yourself afterwards, you have not thought about the problem — you have adopted a conclusion. The distinction shows up in whether the person can still explain the trade-offs when the AI is no longer in the room.

Asking questions that fish for agreement. "Don't you think this is a good approach?" gets you a yes. That yes then feels like external validation, which is worse than no validation, because you now believe you have checked. The question has already embedded the desired answer, so the response adds no new information.

Stopping at the first plausible answer. The initial response is a starting point. The value is in the third exchange, after you have pushed back and asked why and introduced the constraint you left out of the first description. Most of the useful friction appears only after the first smooth reply has been challenged.

Atrophy. This is the long-term risk and it is worth taking seriously. Thinking is a skill maintained by practice. If every difficult problem gets handed off at the first sign of friction, the capacity degrades. The mitigation is straightforward: form your own view first, then bring it to be tested. That ordering preserves the practice and gets you the check.

The Test

A simple way to know whether you are using this well.

After a thinking session, can you explain the reasoning to someone else, in your own words, and defend it against a question you did not anticipate?

If yes, it worked — the tool sharpened something that was yours. If you can only repeat the conclusion, you have borrowed a position, and you will discover the difference the first time someone competent challenges it.

The point is not to think less. It is to have something available at eleven at night that will tell you your favourite idea has a hole in it, without needing to be diplomatic about it.

AI as a Thinking Partner covers the full practice — how these systems actually work, thinking out loud, brainstorming, stress-testing, decision support, problem decomposition, difficult conversations, strategic thinking, and the mistakes that undermine all of it.