Psychological safety: the precondition every wellbeing program forgets

A senior engineer once told me the most expensive bug of his career was one he saw coming. During a code review he noticed a config change would behave strangely under a specific load pattern. He almost said so. Then he remembered the last time he had flagged something in that team’s review — the impatient sigh, the “we’ve already tested this,” the sense that raising a concern counted as the same thing as slowing everyone down. So he approved the change and stayed quiet. The outage cost the company a weekend and very nearly a contract.

Nobody in the postmortem asked the only question that mattered. They asked “how did this slip through review?” when the real question was “why did the person who caught it decide it was safer to say nothing?”

That decision — made in a second, mostly below the level of conscious thought — is what this essay is about. Psychological safety is the shared sense that it is safe to speak up, to disagree, to admit you do not understand, or to disclose that you are struggling, without being humiliated or quietly penalised for it. It is not a perk. It is the precondition for almost everything else a company says it wants from its people, and it is failing in more teams than any org chart would ever reveal.

The term has a real foundation

“Psychological safety” gets thrown around loosely enough that it can sound like a soft slogan, so it is worth grounding. The concept was given rigour by Harvard’s Amy Edmondson, whose research reframed it as a measurable property of teams rather than a vague mood. Her early work produced a finding that still unsettles people: in hospital units, the better teams appeared to make more errors, not fewer. The twist was that they were not making more mistakes — they were reporting more of them. The weaker teams had a similar error rate and simply buried it. Safety did not lower the count of mistakes. It surfaced them, which is how they got fixed.

That distinction is the whole game. A psychologically safe team is not one where nothing goes wrong; it is one where the things that go wrong become visible early enough to be cheap. For engineers and data teams this should land immediately, because our entire reliability practice — blameless postmortems, near-miss reporting, surfacing a flaky test instead of muting it — is psychological safety wearing an operational hat.

The disclosure gap

The clearest evidence that safety is scarce comes from how rarely people admit they are struggling. The 2025 Workplace Mental Health Poll from NAMI, the National Alliance on Mental Illness in the United States, found that only about 13 percent of employees had actually told a manager their mental health was suffering. Set that against the wider picture of strain — Mind Share Partners’ 2025 report puts the share of US workers reporting at least some burnout symptoms near three-quarters — and the gap is stark. The struggle is widespread; the disclosure is almost absent.

The reason is not a mystery; it is risk. In the same NAMI polling, somewhere between 42 and 46 percent of workers said they feared that disclosing a mental-health challenge would damage their career. That fear is not irrational paranoia, and it is not evenly distributed. Deloitte’s Women @ Work 2025 study, which surveyed 7,500 women across 15 countries, found that nearly 90 percent of women believed a manager would think negatively of them if they disclosed a mental-health challenge. When nine in ten people in a group expect to be judged for honesty, silence is not a personal failing. It is a sensible response to a read of the environment.

A caveat worth saying out loud: these are separate surveys with different samples and years, so the bars do not stack into one clean pipeline. Treat them as directional. But the direction is unambiguous and survives the messiness: a large majority of people carry strain they never report, and a near-majority can name exactly why.

What an unsafe team actually looks like

The trouble with measuring safety is that the worst cultures look calm from the outside. Meetings are quiet, reviews are fast, nobody complains. Leaders mistake that quiet for health, when often it is the silence of people who have decided that speaking is not worth the cost. People-management writers in 2025 reached for the phrase “culture rot” to describe this state — slow structural decay underneath a surface that still appears to function.

The markers are specific, and once you know them you start to see them everywhere. In an unsafe team, questions get read as ignorance, so people stop asking and start guessing. Mistakes get traced to a person rather than a process, so the next one gets hidden instead of flagged. Disagreement with the senior voice in the room is treated as friction rather than information, so the room quietly converges on whatever that person already thought. Bad news travels slowly upward while good news travels fast, so leadership’s picture of reality is systematically rosier than the truth. And the tell engineers will recognise: bugs and risks surface late, in production, instead of early, in review — because raising them early carried a social tax that raising them late, when they are undeniable, does not.

A safe team inverts each of these. The contrast is worth seeing side by side, because the behaviours are concrete and observable, not abstract feelings.

The point of the comparison is not that safe teams are nicer. It is that they are better informed. Every behaviour in the right column moves information toward the people who can act on it, faster. Every behaviour on the left throttles that flow. A team that punishes the messenger does not stop the bad thing from happening; it only stops itself from hearing about it in time.

Why this is the precondition, not one initiative among many

Here is the part the wellbeing budget tends to miss. Most corporate wellbeing spending — the meditation subscriptions, the resilience webinars, the mental-health awareness weeks — rests on an assumption that people will engage with it honestly. An employee assistance programme only helps if someone will admit they need it. A “talk to your manager about your workload” policy only works if talking to your manager is safe. A survey of how people really feel only yields the truth if answering truthfully carries no risk. Safety is the layer underneath all of it, and when that layer is missing, the programs above it do not merely underperform. They become theatre — a visible signal that the company cares, performed for an audience that has privately concluded otherwise.

You can watch the same dynamic in the trust data. Gallup has repeatedly documented a perception gap on feedback: around half of managers say they give their people meaningful feedback every week, while only about one in five employees report receiving it. That gap is not mostly about anyone lying — it is a broken two-way channel. The manager genuinely believes they are communicating; the message is not landing as feedback because the relationship is not safe enough for it to register as anything other than evaluation. When the routine exchange of “here is how it is going” misfires by that margin, the far more delicate exchange — “I am not okay,” or “I think we are about to make a mistake” — has no chance at all.

There is a hard-nosed reading here too, not only a humane one. Edmondson’s later work tied psychological safety directly to learning and innovation: teams that cannot voice half-formed ideas cannot iterate on them, and teams that punish failed experiments stop running experiments. For an industry that lives on iteration — ship, measure, learn, correct — a culture where people hide what is not working is not a wellbeing problem with a side of inefficiency. It is a drag on the core loop itself. The quiet engineer who approves the bad config is the same dynamic that, scaled up, produces a product nobody dares tell the founder is failing.

It is built from the manager up

The uncomfortable, and oddly liberating, truth is that psychological safety is local. It is not set by a values poster or a company-wide policy. It is set, team by team, by how the most senior person in each room reacts in the small moments — which means it is largely within a single manager’s power to build or destroy.

The mechanism is reaction, not declaration. Saying “my door is always open” costs nothing and signals nothing. What signals everything is the first time someone walks through that door with bad news and watches what happens to the manager’s face. If a junior admits they do not understand the architecture and the response is a patient explanation, the room just learned that not-knowing is survivable. If someone ships a bug and the postmortem hunts for the gap in the process rather than the flaw in the person, the team learned that surfacing the next bug early is safe. Each of these is a single data point, and people collect them relentlessly, because the cost of misreading the environment is high.

For engineering and data teams the levers are concrete. Run blameless postmortems and mean it — the moment a retro produces a scapegoat, you have taught everyone watching to bury the next incident. Make it normal for the most experienced person in the review to say “I don’t know, let me check,” because nothing licenses a junior to ask a question faster than watching a principal engineer ask one first. Reward the person who surfaces a risk even when it turns out to be nothing — you want a thousand cheap false alarms over one silent real one. And stop conflating two things managers run together constantly: high standards and safety are not opposites. Edmondson’s framing is that you want both at once — demanding work in an environment where it is safe to admit you are struggling with it. The failure mode is not high standards. It is high standards plus a culture where admitting difficulty is fatal, which simply produces people who fake competence until they break.

The cheapest investment nobody books

There is a strange economics to all of this. The wellbeing initiatives companies pay real money for — apps, platforms, retreats — are the ones with the weakest evidence behind them. Psychological safety costs almost nothing in dollars; it is built out of how a manager handles the next moment of bad news, the next naive question, the next disagreement. And yet it is the most expensive thing to fake and the hardest to retrofit, because it is made of accumulated trust, and trust is earned one consistent reaction at a time and lost in a single bad one.

The engineer who stayed quiet during that review was not a coward and was not disengaged. He was an excellent reader of his environment, responding rationally to what it had taught him. The failure was upstream of him, in a hundred small moments where the team had learned that catching something and saying so was not worth the friction.

So the question to sit with is not whether your company has a wellbeing program. It almost certainly does. The question is the one nobody asked in that postmortem: when someone on your team sees the thing about to go wrong, do they believe it is safe to say so out loud — and what, in the last month, taught them the answer?