Confidence is not signal
The cost when management can’t tell the difference between competence and conviction
A VP of engineering I met a few years ago described a promotion calibration that bothers her to this day.
Two candidates for staff engineer. One had quietly architected the system that handled 40% of the company’s revenue. She’d identified the scaling bottleneck six months before it became critical, redesigned the data pipeline, and shipped it with zero production incidents. In the calibration meeting, her manager described her work accurately but briefly. The candidate herself, when asked to write a self-assessment, had listed her contributions in factual, understated terms.
The second candidate had led a high-visibility initiative that was, by most objective measures, a moderate success with significant technical debt. But he presented it as a transformative achievement. His self-assessment was confident, detailed, and expansive. In meetings, he spoke with authority about systems he’d touched peripherally. He volunteered opinions on architecture decisions outside his scope. He radiated certainty.
The panel promoted him. Not because they were biased, they believed. Because he “demonstrated staff-level scope and influence.” The first candidate was told she needed to “increase her visibility.”
The VP told me: “We promoted confidence. We thought we were promoting competence. They’re not the same thing, and our system couldn’t tell the difference.”
The mechanism has a name
In 1999, psychologists David Dunning and Justin Kruger published a study at Cornell that documented what’s now called the Dunning-Kruger effect: people with low ability in a specific domain tend to dramatically overestimate their competence in that domain. The mechanism is elegant and vicious. The skills you need to produce correct work are the same skills you need to recognize correct work. If you lack the competence, you also lack the metacognitive ability to see that you lack it.
Dunning and Kruger found that participants in the bottom 25% of performance estimated themselves near the 60th percentile. They weren’t being arrogant in the usual sense. They genuinely couldn’t perceive the gap between what they’d produced and what good looked like. The incompetence was self-concealing.
The less-discussed second finding is the mirror image: highly competent people tend to underestimate their ability, assuming that what comes easily to them must come easily to everyone. They discount their own expertise because they can see how much they still don’t know. The more you learn, the more the landscape of your ignorance becomes visible.
Put both findings together and you get a prediction that most engineering leaders will recognize immediately: in any room where competence isn’t externally measured, the least qualified person will often be the most confident, and the most qualified person will often be the most restrained.
Where this breaks organizations
The Dunning-Kruger effect is usually discussed as a curiosity about individual psychology. That’s interesting but insufficient. The real damage happens when organizational systems amplify the effect instead of correcting for it.
Promotion decisions. Most engineering organizations evaluate candidates partly on “scope” and “influence,” which are proxies for the ability to operate at a higher level. But scope and influence are assessed through artifacts that favor confidence over competence: self-assessments, calibration narratives, presentation in meetings, visibility of contribution. An engineer who overstates their scope looks like a strong candidate. An engineer who accurately describes their contributions looks like they’re “not quite there yet.” The system rewards the inflation and penalizes the precision.
I wrote about this dynamic in the context of impostor feelings after promotion: top performers often underestimate their readiness because their internal evaluation system is calibrated to what they haven’t done yet rather than what they have. Dunning-Kruger explains the other side. The person who doesn’t feel impostor feelings isn’t necessarily more qualified. They may simply lack the metacognitive resolution to see the gap.
The overstepping peer. The pattern I described in the peer who nods and then goes around you is often powered by Dunning-Kruger. The colleague who designs a project in your scope without recognizing that it duplicates existing work, who redirects resources to their own initiative without understanding why those resources were allocated the way they were, who pitches confidently to senior leadership on a topic they understand superficially: this person is not always acting from malice or status threat. Sometimes they genuinely believe they’re contributing because they can’t see what they don’t know about the domain they’re encroaching on.
This is harder to manage than territorial behavior because the person is sincere. They’re not circumventing your authority strategically. They’re doing it because their internal model of the problem is missing the dimensions they’d need to recognize that your approach already accounts for what they’re proposing. The confidence isn’t a mask. It’s a symptom of the deficit.
Hiring panels. Interview confidence correlates weakly with actual job performance, but it correlates strongly with interviewer impression. Research on interview validity has repeatedly shown that unstructured interviews are poor predictors of performance, in part because they allow Dunning-Kruger dynamics to operate unchecked. A candidate who speaks fluently about a domain they understand at a surface level can outperform, in interview impression, a candidate who speaks carefully about the same domain because they understand its actual complexity. The fluent candidate sounds like they know what they’re talking about. The careful candidate sounds like they’re hedging. The panel reads confidence as competence because the system provides no other signal.
Technical decision-making. In design reviews, architecture discussions, and incident postmortems, the same dynamic plays out. The person who speaks first and most confidently often sets the frame for the conversation, even when they understand the problem least deeply. The person with the most expertise may wait, qualify, and hedge, because they can see the tradeoffs. Dunning himself noted that overconfidence can be advantageous in execution but harmful in planning, because the overconfident person ignores risks that the informed person can see. In a design review, the overconfident voice can set a direction that the knowledgeable engineer then has to spend weeks correcting.
Why the system amplifies instead of corrects
Organizations should, in theory, have mechanisms that make competence legible independent of confidence: code reviews, performance metrics, architectural outcomes, production reliability data. But in practice, most evaluation systems blend objective and subjective signals, and the subjective ones are where Dunning-Kruger does its work.
Calibration meetings rely heavily on narrative. Self-assessments reward self-promotion. Promotion panels weight “leadership presence” and “communication,” which are partly valid criteria and partly proxies for confidence. Performance reviews include peer feedback that can reflect how someone is perceived rather than what they’ve actually produced.
The result is a system that systematically advantages the overconfident and disadvantages the accurately self-aware. And because the accurately self-aware tend to be your most competent people, the system is selecting against exactly the trait it should be selecting for.
This is where the Dunning-Kruger effect intersects with the likability tax and the gender dynamics of evaluation. Women in technical roles are socialized to understate their contributions. Research on gendered self-assessment consistently shows that women rate themselves lower than equally performing men on the same tasks. This isn’t impostor syndrome. It’s accurate self-assessment in a system that rewards inaccurate self-assessment. The Dunning-Kruger effect doesn’t just describe individual cognition. It describes which individuals an organization’s evaluation system will reliably favor and which it will reliably overlook.
The control surface
You can’t fix the Dunning-Kruger effect in individuals. It’s a feature of human metacognition, not a character flaw. But you can build systems that don’t amplify it.
Separate the artifact from the narrative. In promotion decisions, require that the candidate’s work be evaluated through artifacts before the narrative is considered. What shipped? What was the measured outcome? What decisions were made and what were the results? If the artifacts demonstrate staff-level scope, the narrative is confirmation. If the artifacts don’t, no amount of confident self-assessment should fill the gap.
Structure your interviews. The single most effective defense against Dunning-Kruger in hiring is structured interviewing: same questions, same rubric, scored independently before discussion. Unstructured interviews let confidence dominate. Structured interviews force the panel to evaluate what the candidate actually said against a defined standard. Research consistently shows that structured interviews predict job performance significantly better than unstructured ones.
Watch for the quiet expert. In any team, there is probably someone who knows more than they say. They’re not self-promoting because their metacognitive sophistication makes them aware of how much they still don’t know. That awareness is an asset, not a liability. If your evaluation system penalizes it, your evaluation system has a false-negative problem. Actively seek out the understaters. Ask them to describe their work in detail. You’ll often find that the person who says “I just cleaned up the data pipeline” actually redesigned the architecture.
Require specificity in calibration. When someone in a calibration meeting says a candidate “demonstrates broad influence,” ask: where, specifically? What decision did they change? What outcome resulted? Confident generalities are where Dunning-Kruger hides. Specificity is the disinfectant. If the influence can be described in concrete terms, it’s real. If it can only be described in confident abstractions, it may be the effect talking.
Normalize accurate self-assessment. In most organizations, the engineer who says “I’m not sure I’m ready for staff” is told they lack confidence. The engineer who says “I’m absolutely ready for staff” is told they’re showing initiative. This is exactly backward. Build a culture where saying “here’s what I did well and here’s where I’m still growing” is treated as a sign of sophistication, not weakness. The goal is a promotion system that rewards self-knowledge, not self-promotion.
Signals
You’ll know your system is correcting for Dunning-Kruger when promotions start surprising people in the right direction: when the quiet engineer who shipped the critical system gets recognized without having to campaign for it. When calibration meetings spend less time on narrative and more time on artifacts. When the person who speaks most confidently in the design review isn’t automatically assumed to be the most informed. When “I don’t know” starts being treated as a sign of expertise rather than a gap in it.
Confidence is not signal. In a system that can’t distinguish between competence and conviction, the most capable people are penalized for their own accuracy, and the least capable are rewarded for their inability to see what they’re missing. That’s not a hiring problem or a promotion problem or a calibration problem. It’s a measurement problem. And the first step to fixing it is admitting that your current instruments are measuring the wrong thing.