A team that laughs a lot isn’t more creative because of it. The causality runs the other way, and before you reverse it, it’s worth understanding why every piece of advice along the lines of “loosen things up a bit” fails.
I was once in a meeting where one man talked. Everyone else answered when questioned, and nodded. No jokes, but also no criticism and, as it turned out, nothing of substance. Everyone in that room knew things about the project being discussed. None of them were said.
In another team we laughed a lot and delivered. At what? At things we had no control over: deadlines, the absurdities in briefs. And when we screwed something up, we laughed at ourselves.
Nale Lehmann-Willenbrock and Joseph Allen recorded 54 production teams at two German companies and coded their meetings second by second. The result, published in the Journal of Applied Psychology in 2014, contains one finding that should change how we think about humour at work. Individual jokes correlated with nothing. Laughter alone, nothing either. Statistically: zero. What mattered were patterns - sequences where a joke triggers laughter, which triggers another joke. Those patterns related to team performance at β = .33 immediately and β = .35 two years later. And they triggered a specific kind of communication: procedural suggestions, summaries, mutual praise, and at a longer lag, new ideas and solutions.
But there’s a condition that eats everything else. In teams with a high climate of job insecurity, the relationship vanished. People still joked. The jokes stopped meaning anything.
That’s the core of it. Humour doesn’t create a safe environment; it’s what a safe environment produces and then amplifies. If people on your team joke about things that are genuinely uncomfortable, it isn’t because you have a good culture of laughter. It’s because you have a culture where you can risk something, and a joke is simply the cheapest risk available.
Take a joke apart. For something to be funny, it has to violate a norm - otherwise it’s boring. But that violation has to be safe - otherwise it’s aggression or horror. And both things have to register at once - otherwise it’s just awkward.
Three elements, all necessary. Remove any one and the joke dies.
Now read Amy Edmondson’s definition of psychological safety. It’s the belief that I can take an interpersonal risk and not be punished for it. Ask a stupid question. Admit a mistake. Say I don’t understand.
Same structure. Literally the same. A norm violation plus a guarantee that the violation is safe.
That’s why humour is such a good indicator. A joke is the smallest possible unit of a safety test. It costs a second, requires no courage, needs no preparation. Someone throws out something that mildly violates a norm and watches what happens. If the team laughs, everyone just received the information that here you can. If silence falls, they received information too.
People run this test constantly, without knowing it. Before anyone says at a meeting “I think this project is going nowhere,” they’ve checked twenty times whether it’s permitted here to say something that doesn’t match the line. With a joke. Because a joke can be withdrawn. “Relax, I was kidding.”
Which is why every “culture of fun” initiative is causality in reverse. One telecoms company rolled out a fun-culture programme and earned itself a study describing the effect: cynicism and resistance. Employees read the programme as patronising and inauthentic. Not because they don’t like having fun, but because fun imposed from above isn’t a norm violation. It’s the norm. A norm violated by the person who sets the norm is just another instruction.
A ping-pong table doesn’t create safety. A ping-pong table is what you do when you don’t want to deal with what’s actually bothering people.
Who gets to make the joke
Everything I’ve written so far leads straight to the advice “joke more.” And that advice is worthless, and I’ll explain why.
Bitterly, Brooks and Schweitzer ran 8 experiments. The mechanism they found is brutally simple. A successful joke signals two things at once: confidence and competence. That raises status. A failed joke signals confidence paired with low competence. And that particular combination lowers status - more than if you’d said nothing at all.
A joke isn’t free. It’s a bet.
The worse part came two years later. Jonathan Evans and his team recorded four versions of the same presentation by a retail store manager. Identical script, identical jokes, delivered once by an actor and once by an actress. Participants watched one version and evaluated it. When the man joked, it read as functional - status up, better ratings on performance and leadership. When the woman joked, it read as disruptive. Status down.
The authors named it directly: another gender-based inequity in the workplace.
Someone will point out that we do see funny women who benefit from it. Sure - on stage. A woman standing on a stage has a platform by definition, because someone invited her, introduced her and handed her a microphone. The status came before the joke. A woman in a weekly team meeting has nothing but herself, and her joke is supposed to produce the status in the first place.
So humour works in favour of those who already have status, and against those who don’t. It isn’t a tool for gaining position; it’s a privilege of having it.
What follows practically? If you’re a leader and your team rarely jokes, the problem isn’t that they’re tense. The problem is they calculated the risk and concluded it doesn’t pay. And if you’re the only one joking and everyone laughs, that isn’t a culture of humour. That’s a court.
The manor where only the master jokes
“When the master is in a good mood, he’ll praise you. When he’s furious, he’ll scold you, shout at you, throw you out of your job.”
That’s how Jacek Santorski describes the modern Polish company. Read that sentence again and notice where the humour sits.
The humour belongs to the master. It’s his state, his prerogative, his weather. The master’s mood is the independent variable, and everyone else is a dependent one. This is the exact opposite of what Lehmann-Willenbrock and Allen found, where humour was a pattern - something that circulates, returns, bounces back. An exchange. In the manor, humour doesn’t circulate. It trickles down.
Santorski didn’t invent this diagnosis. Behind it stands the work of Janusz Hryniewicz, who studied Polish organisations and concluded that around 75% of them - corporations, small firms, public offices - run on the culture of a medieval manor. Communication flows one way only, downward. Unlimited power and the master’s distance on one side; obedience, servility and a double ethic on the other. One thing is said in front of the master, another behind the barn.
More than once I’ve been in a meeting where the conversation changed completely after the CEO left the room. That alone isn’t a diagnosis of the manor - back channels form in every team and you have to allow for them. It’s a natural release valve. The question is what separates those two conversations. If the second one is simply looser, everything is fine. If the second one contains substance that couldn’t be said in the first, you don’t have a release valve. You have a shadow system.
This ties together everything above. Bitterly showed a joke is a bet on status. Evans showed the stakes differ depending on who you are. Hryniewicz describes a structure where 75% of Polish organisations have exactly one player who can afford that bet.
Polish culture has a figure who understood this 500 years ago. Stańczyk. The court jester, the only man at court licensed to tell the king the truth - because truth delivered as a joke wasn’t an assassination attempt. Matejko painted him sitting alone while the rest of the court dances in the next hall. The jester is the only one who knows.
I know how it works because I’ve played that role many times. An outside consultant gets Stańczyk’s licence for free - he can say things an employee would pay for with his career, because in a week he’ll be gone anyway. It’s convenient. It’s also proof that something is broken: if you have to hire someone from outside to say things half the room already knows, the problem isn’t a lack of knowledge.
It’s worth asking this about your own company: who here holds Stańczyk’s licence? Who can tell the board something the board doesn’t want to hear and walk away intact?
(Cabaret in communist Poland worked on the same principle. It wasn’t entertainment; it was the only available channel for truth. When the only way to say something true is as a joke, that says nothing about a society’s sense of humour. It says everything about the state of the other channels.)
The people who ruin the profile
You know that moment in a workshop when someone is assigned the devil’s advocate role and spends ten minutes inventing objections he doesn’t believe. Everyone at the table knows he doesn’t believe them. Then we return to the original idea with a sense that it’s been tested.
It hasn’t.
Charlan Nemeth at Berkeley showed in 2001 that an appointed devil’s advocate doesn’t stimulate divergent thinking the way an authentic dissenter does. The reason is obvious once you hear it: since everyone knows he’s only playing a role, nobody has to genuinely reconsider their position. The objection doesn’t come from someone who believes it. It comes from a function. Nemeth points to one more thing: a real dissenter takes a risk, because he actually stands behind his position. A devil’s advocate risks nothing, because he was given instructions. And everyone at the table knows it.
The consequence is uncomfortable for anyone who likes procedural solutions. You can’t install criticism. There is no process, workshop or ritual that will manufacture genuine dissent out of people who agree. You have to hire someone who disagrees. Actually, not as a role.
And here comes something that sounds like a contradiction of that paragraph - but isn’t.
I once consulted on the recruitment of two people and after a few minutes of conversation I recommended against hiring them. They were hired. Within a few months nearly everyone on the team saw the same thing: morale down, project quality down, engagement down. The moment I had the authority, I released both from their duty to work, effective immediately.
Michael Housman and Dylan Minor analysed data on nearly 60 000 workers across 11 firms. The finding: avoiding one toxic worker is worth more than hiring a star from the top 1% of performers. Roughly two to one. Toxicity spreads - exposure to a toxic colleague measurably raises the odds that you’ll start behaving that way yourself.
The most insidious part of their findings is that toxic workers often look productive. They’re fast. They deliver. The manager looks at the numbers and sees a star.
So where’s the line? A dissenter and a toxic worker look similar from outside: both are inconvenient, both say things nobody wants to hear, both ruin the profile.
The line doesn’t run along opinions. It runs along respect for people. A dissenter attacks the idea and respects the person. A toxic worker attacks the person and uses the idea as a pretext. A dissenter wants the decision to be better. A toxic worker wants to win.
That distinction is easy to state and hard to apply, especially when you’re the one the dissenter has just attacked. But it’s the only one we have, and any other - based on tone, or on how pleasant someone is - will function in practice as a filter for conformity.
One more thing. A single such person is usually not enough. Dropped into a team, a lone dissenter becomes a token, an oddball everyone tolerates and nobody listens to. The dynamic shifts only at critical mass, somewhere around 30%. If you hired one person who thinks differently and she leaves after a year, frustrated, it isn’t because she didn’t fit. One person isn’t diversity - it’s an experiment with a sample size of one, run on someone who didn’t volunteer.
(I’ve been that person.)
The broken filter
Two percent. That’s how many people on the autism spectrum are economically active in Poland, according to research by the Polish Economic Institute and the JiM Foundation from 2022. The EU average is around 10%. Five times fewer. Among people with disabilities in Poland overall, employment stands at 26.4% - 13 times higher than among people on the spectrum.
Neurodivergent people make up 15% to 20% of the population. The number of pupils diagnosed with autism grows by around 20% a year. 82% of Polish small and medium enterprises report difficulty finding workers.
Put those numbers together. A market desperate for people systematically filters out 15% of the population.
This is usually where the productivity argument shows up. That companies employing neurodivergent people post X% better results. I won’t use it, because those numbers come from company materials, circulate in inconsistent versions and have no peer-reviewed publication behind them. The argument isn’t needed anyway, because the real one is more interesting.
The filter is broken and we know exactly where
Standard recruitment measures things unrelated to the work. The behavioural interview rewards the ability to narrate yourself in a story structure. The small talk at the start rewards fluency in conversation about nothing. Eye contact. “Culture fit” - which means, let’s call it what it is, whether we’d be comfortable having lunch with this person. None of these is the job. All of them are a filter for neurotypicality.
And the cost on the other side? The Job Accommodation Network, an agency of the US Department of Labor, collected data from 1 425 employers. 61% of accommodations cost nothing. 33% were a one-time cost with a median of 300 dollars.
Three hundred dollars. For access to a talent pool your competitors won’t touch, in a market where 82% of firms can’t find anyone.
The best-documented example I know doesn’t come from any corporation. GCHQ, British intelligence, deliberately recruits dyslexics. According to the agency, a GCHQ apprentice is three to four times more likely to be dyslexic than on other schemes. The reason is purely operational. Jo Cavan, GCHQ’s Director of Strategy, puts it plainly: they’re looking for people who can spot something out of place in a bigger picture, who have strong visual awareness and can catch anomalies. When you’re sifting vast quantities of data from many sources to stop a terrorist attack, pattern recognition is key. Cavan calls it exactly what it is: mission critical.
This isn’t a CSR programme but rather an intelligence agency that needs people who see what isn’t visible.
The price of admission
And here humour comes back, in a way that’s inconvenient for everything I’ve written above.
Research on humour in people on the autism spectrum says one thing with certainty: the sense of humour is absolutely there, it just has a different profile. Less affiliative humour - the kind that builds bonds with a group. A preference for simpler structures. On the rest, researchers argue - some claim absurdist humour lands worse, others that it lands better than in neurotypicals. The dispute isn’t settled and let’s not pretend otherwise.
But the practical conclusion points clearly, and unpleasantly. A culture built on fast, contextual banter - on irony, sarcasm, status allusions, on jokes that require knowing the previous three jokes - is an exclusionary culture. Not because it’s malicious, but because it demands a very specific kind of social processing as the price of admission.
So we have a contradiction. Humour is an indicator of safety, and at the same time the particular kind of humour that most often performs that function in teams is a barrier to entry for the very people we want inside.
I don’t have an elegant solution and I’m not going to pretend I do. I only have an observation: a culture where you have to keep up with the banter to be taken seriously is not a safe culture. It’s a culture with an entrance exam, just better dressed. The real test isn’t whether people laugh at your place. The real test is whether someone who doesn’t get the joke can say so.
The mistake that’s an investment
One element remains, and I have to start by admitting something.
We all know the cost-of-defect curve. A bug caught at design costs one unit, in development ten, in production a hundred. Anyone who’s sat through a quality training knows it. It’s attributed to IBM. And it has no source - Laurent Bossavit dug through the citation chain and arrived at internal training notes with no data behind them. Folklore, repeated for forty years, because it sounds reasonable and everyone cites everyone else.
That cost me the most convenient argument in this essay, and I have to build it differently. Without the number.
In “Right Kind of Wrong,” Amy Edmondson splits failures into three types, and that distinction does all the work. Basic failures: simple, in known territory, single-cause - inattention, haste, overconfidence. You don’t celebrate those; you prevent them. With a checklist. Complex failures: many causes, a complex system, warning signals someone ignored. You don’t want those, but they’re inevitable and they need taking apart. And intelligent failures: new territory, knowledge-based, aimed at a specific goal - and small.
Pay attention to that last word, because it holds the whole answer to the question I can no longer back with a number. Being small isn’t a benefit of an intelligent failure. It’s its definition. If your failure was big, it wasn’t intelligent - not because it turned out badly, but because you let it grow before you tested the assumption.
That’s why a workshop, a paper prototype, a quick study with five people, an ugly alpha make sense. Not because there’s some multiplier, but because at that stage nothing has been built on the wrong assumption yet. Every week of delay is more decisions someone made resting on something that isn’t true. You don’t pay for the mistake itself. You pay for everything built on top of it.
Edmondson began her entire body of work on psychological safety with a finding that initially looked like a data error. She studied medical teams and found that better teams reported more errors. The result was absurd. Only later did it become clear: better teams weren’t making more errors. They were reporting them.
A team that doesn’t report errors isn’t a team without errors. It’s a team where errors are expensive to admit, so they grow in silence until they’re big enough to reveal themselves. Which is precisely the moment when fixing them costs the most.
And here the whole chain closes.
Psychological safety makes people report errors early. Errors reported early are small, because nobody has built anything on them yet. Small errors are intelligent errors. Intelligent errors are the only known mechanism for learning in new territory.
And humour? Humour is how you know the first link in that chain exists at all. It’s an indicator, not a lever.
Low-maintenance
Companies write in job ads that they’re looking for proactive people. Then they wonder why the proactive people they hired are waiting for instructions six months later, like everyone else.
Researchers on proactivity point to something those job ads never mention. Proactivity is risky by nature. It questions the status quo, its outcome is uncertain, and if it goes wrong, everyone knows who started it. Nobody takes initiative without first calculating what it will cost them if it fails.
So proactive people aren’t low-maintenance because they’re some superior species of employee. They’re low-maintenance when the environment makes initiative cheap. That same person in the manor will wait for instructions - not out of bad will, but because they did the math. That’s the difference between a person and the conditions you’ve placed them in, and most companies consistently confuse the two.
In every company I walk into, there are people who’ve gone dim. It’s easy to say they lack ambition. Usually it’s the opposite: they once tried to change something and got slapped down for it. What you’re seeing isn’t their character - it’s a defence mechanism working exactly as it should. For that person to risk it again, you need time and an understanding of what happened to them. You can’t do it with an engagement workshop.
This is also the answer to the question about scaling. Scale a structure where every decision has to travel upward, and coordination costs will eat you alive. Scale a structure where people make decisions themselves, and you grow. But people only make decisions themselves when a mistake doesn’t end in a public execution. Quality at scale isn’t a function of process. It’s a function of what a mistake costs.
A joke, a dissenter, a reported error and a proactive initiative are four forms of the same thing: someone risking a norm violation because they calculated it won’t come back at them. They differ only in price. The joke is cheapest, which is why it comes first and why it’s the indicator. A reported error costs more. Contradicting your boss costs most.
If the first one isn’t there on your team, don’t count on the other three.
You can’t run a workshop on humour. You can’t appoint a devil’s advocate. You can’t buy proactivity with a job ad. All of these are symptoms of a single variable, and that’s the only thing you can act on.
The silence on your team isn’t a trait of their character. It’s a result you produced, and they did the math correctly.
So the question isn’t how to loosen things up. The question is: when was the last time someone made a joke about something involving you?
If you can’t remember, that isn’t a compliment.