The technological impostor syndrome. We all have it at once

Today it's time to talk about a feeling almost all of us who use AI at work have and almost nobody owns up to: signing something you didn't entirely do. My thesis is that the old impostor syndrome, which used to be a thing for a few, is becoming the normal state of an office. But is that a syndrome, or is it simply lucidity? That's where my doubts begin. My history with AI goes back to 2005, with a language called Prolog, and even so I don't know whether what's changed since then is the tool or us.

A clinical picture from 1978

In 1978 Pauline Clance and Suzanne Imes published in Psychotherapy. Theory, Research and Practice an article titled The Imposter Phenomenon in High Achieving Women. They described women with outstanding CVs who lived convinced their success was a fraud: they'd been lucky, they'd fooled the panel, someone would catch on eventually. In the original formulation the phenomenon was individual and it was a minority one. The impostor stood out against a background of people who did deserve what they had.

Forty years later, the label has been stretched so far it's hard to know what it measures. A systematic review by Dena Bravata and colleagues in the Journal of General Internal Medicine (2020) brought together 62 studies with more than 14,000 participants and found prevalences ranging from 9% to 82%, depending on the questionnaire and the cut-off point. When one and the same concept produces figures that far apart, the problem isn't only in the people being surveyed.

What programmers and CEOs say

With that caution, recent data point to something widespread. Paloma Guenes, Margaret-Anne Storey and three other authors surveyed 624 software engineers for a paper presented at ICSE-SEIS 2024, in Lisbon. 52.7% experienced the phenomenon with high frequency or intensity; women more (60.6%) than men (48.8%). It's the sector generative AI got into earliest and deepest.

Korn Ferry, in its Workforce 2024 report, surveyed more than 10,000 employees, among them 400 CEOs. 71% of the US ones said they felt symptoms of the syndrome, against 33% of professionals just starting out. It's a consultancy survey, not peer-reviewed, and the same survey records that 85% of those CEOs consider themselves fully competent. Feeling like an impostor and feeling capable, at the same time. That alone should make us suspicious of the word.

Where AI fits in

My intuition is that AI has added a new variant: functional opacity. You operate an instrument whose inner workings you don't know, in an environment where everybody makes a show of knowing how it works, and with it you produce results that carry your signature.

Three scenes, all ordinary in 2026. A copywriter hands in a text a model generated and in which she changed five words; if her editor asked her to justify each paragraph, she couldn't. A programmer accepts a function proposed by a coding agent, it passes the tests and goes to production; when it fails three months from now, will he know how to debug it? A consultant delivers a presentation with charts and a summary made by the machine, signs it and bills for it. For what, exactly?

A finding from METR, an independent evaluation lab, shows how slippery self-perception is on this ground. In a randomised trial with 16 veteran open-source programmers and 246 real tasks, between February and June 2025, those using AI took 19% longer. They believed they'd been 20% faster. Careful, though: METR declared that result superseded in February 2026: with newer tools the effect seems to have flipped, and it couldn't measure it properly because between 30% and 50% of the programmers refused to do part of their work without AI. Hang on. That refusal is data too. Once you no longer want to work without the tool, the question of what you can do on your own gets uncomfortable.

The expert's paradox, with small print

The original version of this article leaned heavily on the Dunning-Kruger effect. Justin Kruger and David Dunning published Unskilled and Unaware of It in 1999: the least competent overestimate their competence and the most competent tend to underestimate it, because they can see the map of what they don't know more clearly. Applied to AI, it came out neat: whoever doesn't understand a language model uses it without unease; whoever understands its failure modes uses it with a nagging suspicion.

The small print is that the effect is under dispute. Gilles Gignac and Marcin Zajenkowski argued in Intelligence (2020) that a good part of it is a statistical artefact, regression to the mean mixed with a general bias towards believing ourselves above average. A 2023 reply in the same journal maintains that, with the right statistical methods, the effect still shows up. The debate is open, and the coffee-mug version doing the rounds on social media hasn't been won by anybody. I'll stick with the modest version: knowing more about a field makes what you don't know more visible. That's enough for the argument, and there's no need to inflate it.

The machine's advocate

There's a serious objection to all of the above and it deserves its space, because people who think well hold it.

It goes like this: nobody has ever fully understood their tools. The programmer of 1995 didn't know what the compiler did with their code; the one in 2015 didn't know what the library they imported did; the doctor reads an MRI without being able to reconstruct the physics of the magnet. All of technical civilisation runs on layers of abstraction that delegate trust. Calling that imposture is confusing the division of labour with fraud. What you get paid for isn't having made every piece, but answering for the result.

The objection has empirical backing too. Ruchika Tulshyan and Jodi-Ann Burey published in Harvard Business Review (2021) a widely read piece, Stop Telling Women They Have Imposter Syndrome, arguing that the label individualises problems that belong to the environment. And a 2026 study in the International Journal of Human–Computer Interaction (Cipriani, Menicucci, Greco and Grassini), which built a scale to measure shame, guilt and imposture linked to AI use, found that the feeling of being an impostor goes down with frequency of use. People who use it a lot feel less like impostors. Read that way, the unease would be transitional: whoever's halfway there feels it, and it passes with habit, the way it passed with the calculator.

My reply, which isn't a complete refutation

I accept half. Layers of abstraction are old and they aren't fraud. The compiler, the library and the MRI have a property the language model lacks: they do the same thing every time and someone can explain why. When they fail, they fail in a locatable way. The opacity of a generative model is of another kind, because the same request produces different results and nobody, not even whoever trained it, can reconstruct why that sentence came out and not another.

And the bit about imposture dropping with use doesn't reassure me. It unsettles me. It could mean we learn to use the tool well, or it could mean we stop noticing the distance between what we sign and what we know how to do. The study is cross-sectional and doesn't distinguish between the two. I suspect a bit of each is going on, and that the second can't be seen.

What stops being pretended

Nicholas Carr, in The Glass Cage (2014), described the problem with airline pilots. The modern pilot doesn't fly less, he flies differently: his expertise is monitoring an automatic system and knowing how to step in when it fails. That expertise only gets trained when the system fails, and the system rarely fails. A hundred hours of flight, ten minutes of learning.

Carried over to the office, the copywriter, the programmer and the consultant no longer produce from scratch: they supervise. Supervision is a real and difficult skill, but it gets little exercise. What the technological impostor syndrome detects, I think, is exactly that crack. Not the feeling of not deserving the achievement, which was Clance and Imes's thing, but the half-awareness that you're not training the very thing you're paid for.

And if everyone on a team feels it at once, the category dissolves. Impostor stops meaning anything when legitimate runs out of a referent. Which isn't liberating. It's a question nobody has asked out loud in the Monday meeting.

Definitions

• Impostor syndrome (impostor phenomenon): pattern described by Pauline Clance and Suzanne Imes (1978) consisting of the persistent conviction, not justified by the facts, that one's own achievements are fraudulent and will be found out.
• Functional opacity: a situation in which someone operates a system without understandable access to its internal logic and depends on the result without being able to inspect the process.
• Dunning-Kruger effect: tendency, described in 1999 and now disputed in its magnitude, whereby people with little competence overestimate their performance and highly competent people underestimate it.
• Deskilling: the process by which a professional skill atrophies or is transformed when it's delegated to an automated system.

References

• Clance, P. R. & Imes, S. — The Imposter Phenomenon in High Achieving Women. Dynamics and therapeutic intervention. Psychotherapy. Theory, Research and Practice 15(3), 241-247 (1978).
• Bravata, D. M. et al. — Prevalence, Predictors, and Treatment of Impostor Syndrome. A Systematic Review. Journal of General Internal Medicine 35, 1252-1275 (2020).
• Guenes, P., Tomaz, R., Kalinowski, M., Baldassarre, M. T. & Storey, M.-A. — Impostor Phenomenon in Software Engineers. ICSE-SEIS 2024, Lisbon. DOI 10.1145/3639475.3640114.
• Korn Ferry — 71% of U.S. CEOs Experience Imposter Syndrome (press release on Workforce 2024, June 2024). Corporate survey.
• METR — Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity (July 2025) and We are Changing our Developer Productivity Experiment Design (24 February 2026).
• Kruger, J. & Dunning, D. — Unskilled and Unaware of It. Journal of Personality and Social Psychology 77(6), 1121-1134 (1999).
• Gignac, G. E. & Zajenkowski, M. — The Dunning-Kruger effect is (mostly) a statistical artefact. Valid approaches to testing the hypothesis with individual differences data. Intelligence 80 (2020).
• Dunkel, C. S. et al. — Reevaluating the Dunning-Kruger effect. A response to and replication of Gignac and Zajenkowski (2020). Intelligence 96 (2023).
• Tulshyan, R. & Burey, J.-A. — Stop Telling Women They Have Imposter Syndrome. Harvard Business Review, 11 February 2021.
• Cipriani, E., Menicucci, D., Greco, A. & Grassini, S. — Emotional Responses to AI Use. Development of the SAG-AI Scale for Shame and Guilt. International Journal of Human–Computer Interaction (2026). DOI 10.1080/10447318.2026.2648803.
• Carr, N. — The Glass Cage. Automation and Us (Norton, 2014).

Further reading

• Vallor, S. — Technology and the Virtues. A Philosophical Guide to a Future Worth Wanting (Oxford University Press, 2016). Her reflection on moral deskilling extends Carr's logic: with automation it isn't only technical skill that atrophies, practical judgement does too.

## You might also like - The illusion of learning - The erosion of one's own judgement - Functional ignorance - The black box as the norm - The extinction of human skills

## Elsewhere - Korn Ferry — 71% of U.S. CEOs Experience Imposter Syndrome - METR — We are Changing our Developer Productivity Experiment Design - Guenes et al. — Impostor Phenomenon in Software Engineers (full text, PUC-Rio)