I have always struggled with overconfidence, mostly my own. I wrote about it in When Did You Last Change Your Mind On Purpose?, and about how radical open-mindedness made me question what I thought I knew, because leaning into my convictions has always been the more dangerous option for me than making peace with what I did not know for sure. I have even been given feedback by one of my team members once that I was asking feedback when I didn’t need to, and perhaps I was overthinking it, even though, by all accounts and if you ask anyone who knows me personally, I would describe myself as someone pretty full of herself and definitely on the confident side.
Yet I came up in an era that rewarded confidence, even the unjustified kind, and that has not changed. At work, on social media, among friends, everyone has an opinion and not too rarely the loudest ones have very little substance to back them up. I have had my fair share of incompetent bosses, so when my friend B, a fellow senior executive, sent me an article from HLP Consulting called The Rise of Competent Incompetence, I found myself nodding along rather vigorously. The one that was “sidemoted”, the one that was “compromised-agreed” in your way, or simply the one that got up by inertia as nobody took the time to question or deal with him. We all met them and if you have not yet, brace as they are everywhere.
The appearance of being good
Bronwen Jones, writing for the Cognitive Work Revolution, uses the phrase competent incompetence to describe people who are not good at the job itself yet are extremely good at producing the appearance of being good at it. Often the register and the words used are the giveaway. Phrases like: “We will leverage our capabilities to build a bridge to the future” or “We must move the needle on progress” or maybe “We are moving to be an AI powered company”. Ask for specifics and you get a committee, a further analysis or a meeting about a meeting. Chris Argyris gave it a pretty ironic yet accurate name: skilled incompetence. First rule? Avoid the conversations that will get you exposed.
Competent incompetence is the performance of certainty that others can eventually see through. With AI as a tool to use, our ability to discern information and learn how to deal with trained and efficient overconfidence is more urgent than ever. I really realised in my workshops that we tend to trust the tool differently than we would a colleague as I covered in “Because Claude Said So, And Other Things I Have Heard In Meetings” a few weeks ago.
The illusion of greater confidence
Moving away from what felt like talking about a boss or two I had in my past, perhaps the most incredible experiment is the one of Cornelia Walther, posted last week in Psychology Today. With a very decent sample of over three thousand people who were asked difficult questions and given the option of answering that they did not know, it was shocking to me to read how many people were willing to take wrong answers for good ones without hesitation.
Then they were all given deliberately incorrect AI advice.
Before listening to the machine, enough people were willing to withhold judgement, as about forty-four per cent did. With AI, their willingness to say they did not know dropped to three per cent. Worrying, I’d say. Accuracy fell from twenty-seven per cent to nine per cent. Confidence more than doubled, from thirty to seventy-six per cent… on advice that was wrong.
More confident yet very much less right. Not the best of trades.
AI agents also seem to score higher on confidence than human agents even when acting the same, which Walther calls an illusion of greater confidence. Carnegie Mellon and Microsoft found that knowledge workers who trusted AI output blindly showed measurable cognitive erosion, while those who stayed sceptical and put effort into critiquing it kept their edge. I titled my keynote “The Intuitive Advantage, Expand Your Edge in The Age of AI” for a reason. Judgement is in dire need of support.
Why this matters more for intuition than for anything else
We have to hold a signal before we can explain it. SPACE, my tool to interrogate what we intuitively sense, exists for that. I have written plenty against the tempo trap, leaving a divergence open in a room, and what a possum taught me about thinking, which is a piece entirely on incubation.
Not knowing is not a gap to be closed as fast as possible but the fertile soil in which intuitive signals get a better chance to show up and maybe, through opportunistic assimilation, become insights.
Which is why the work of Kağan Kırcaburun at Düzce University caught my attention. Looking at how people use AI, he found that one of the things most closely linked to problematic use is a low tolerance of uncertainty, which in some cases can be considered pathological. AI soothes them as it offers fast, certain answers wrapped in total confidence for good measure.
I think the saddest part is that these people, who arguably need their intuition the most, are the ones least able to use it, as they can’t hold a question open for long enough and have the strongest tendency to resort to the machine to get fast, confident, reassuring answers. Which appeal to them, even when those answers are wrong.
An intuitive signal arrives before before you can put it into words. A chatbot is, above all, a certainty machine that is built around predictive use of the most likely next word. It delivers finished language, of the most statistically average type, before any sensing has taken place.
Use it too heavily and your intuition simply has no chance to show up.
Where AI fits
If you are familiar with my work you will know that I value AI and use it daily. It’s one of the top three tools I use to make decisions alongside traditional data and information and intuition.
The issue is timing as well as the intimate one to one relationship we seem to have created with the tool. AI should be interrogated only after you have a chance to get your thoughts around an issue or a task and ideally brainstormed those thoughts with a trusted set of humans.
And you should sit with that thought long enough to have given your intuition a chance to show up.
Assisted intelligence needs assisted judgement. Collective intelligence is always stronger than any single person alone.
What this means for you this week
I was talking to someone very smart today, leader of an AI practice, mentioning to me how hard he finds going back to his original thread of thoughts after Claude got hold of it and recommended all sorts of additive, different or complementary things. My answer was to write down with pen and paper all your key thoughts before you even type the prompt, or even better stick them to a wall. Then consider, compare, add and remove following what your LLM of choice suggested.
The silver lining in Walther’s piece is that when accuracy was rewarded and mistakes had consequences, people became more cautious, followed bad advice less readily, and performed better. We are not all doomed.
So my advice is that if the honest answer is that you do not know, say so out loud. It’s OK. You are stronger because of it. I won some of my best clients and deepened some of my strongest commercial relationships making clear the edges of my and my team’s knowledge. They trusted me tenfold more because of it.
What’s next
The other half of this. If the machine is training the rest of us out of saying we do not know, it is also handing the people who could never say it an audience that never tires of them and never leaves. And the implication it carries for those of us who have spent a year practising radical open-mindedness.
Send it to someone who is confident enough to say “I don’t know” often. Most likely they are the most knowledgeable ones anyway. Or as a subliminal “maybe it’s time to admit you don’t know it?!” to someone who needs just that. They will be able to figure out who is who by themselves… or not ;)
If you are not yet following Intuition Edge, you can do that here: substack.com/@intuitionedge
Published every Monday.


LIVIA, forty-four percent to THREE PERCENT??
That is horrifyingly fascinating. People were actually willing to sit with “I don’t know,” then a machine confidently wandered into the room with the WRONG answer and suddenly almost everybody found religion. Confidence more than doubled while accuracy fell from 27% to 9%.
And I think that’s why your distinction about timing matters so much. You’re not doing the boring “AI bad, throw laptop into sea” argument. You USE it. You’re asking what happens when we invite it into our thinking before we’ve even heard our own thoughts.
Also, “competent incompetence” has unfortunately given me a name for several people I have met.
I will be behaving terribly with this information.
Nicely landed. And I agree with you. My ultimate observation goes to the conclusion that beyond our perception of time and space, there is none. This strata of intelligence is as omnipotent and right-full as anything, has no greater or lesser importance than any other aspect of the whole, including our species, it is foundational, we are just starting to experience it. Seeing the rationalising of it is as fascinating as the tools themselves. I think you are acting as a brilliant sentinel to watch the edges and move us towards health as this dance learns itself. Great article. Your writing is vivid.