I’ve seen more teams damaged by the reply to a problem than by the problem itself. A deploy breaks, an estimate turns out to be wildly wrong, a process is eating someone’s week. Those things are fixable. What’s much harder to fix is what happens when an engineer raises one of them and the person in charge defends, explains or says “well, actually”.
That engineer has just done something useful. They’ve spotted a problem and they’ve taken a small risk to tell you about it, because they want the work to be better. If the reply treats that as an attack, they learn not to do it again. And the next time something’s wrong, you hear about it from the incident report or the resignation letter.
The reply is the moment that counts
Psychological safety is the shared belief that you won’t be punished or embarrassed for speaking up with a question, a concern or a mistake. Amy Edmondson at Harvard defined it, and I’ve written separately about psychological safety in engineering teams, the version a CTO builds day to day.
What I want to focus on here is where it gets built, and in my experience that’s mostly in the few seconds after someone says “this isn’t working”, because that’s when they find out whether it was safe to say. Every one of those moments is a small test, and the team keeps score. In a startup the replies being scored are mostly the founder’s, which is why I think psychological safety in startup culture is the founder’s job.
The damage usually comes from a run of small replies that each made sense at the time. The founder explains why the roadmap is the way it is. The tech lead points out that the flaky test was flagged months ago. The manager says the estimate would have been fine if the requirements hadn’t changed. Each reply is reasonable on its own, and together they teach the team that raising things gets you a defence, not a fix. So people stop.
Why a problem sounds like blame
If you built the thing, a complaint about it lands on you. You hear “the release process is a mess” and part of your brain hears “you made a mess”. So you explain the constraints you were under, which feels like context to you and sounds like a defence to them.
The fix is to separate what was said from what you heard. The engineer said the process wastes their time. They didn’t say you’re bad at your job. Nearly every complaint at work is someone trying to improve the process, the system or the relationship. Hear it as that and the reply gets much easier.
You can still explain. The order is what matters. When you lead with the explanation, the person learns that their job in this conversation is to listen to you. When you lead with thanks and a question, they learn that you wanted to know.
What the research shows
Three bits of research are worth knowing here, because each one changes what you’d actually do.
In 1996, Edmondson studied medication errors on eight hospital units. She expected the better-led units to make fewer mistakes. Instead, the units with stronger leadership and better relationships had higher detected error rates. Her explanation was that people on those units were more willing to report errors, so more of them were found. For you, it means this: when a team starts feeling safe, the number of problems you hear about goes up. That’s the thing working, so don’t panic when it happens.
The sociologist Ron Westrum sorted organisations by how information flows through them, and one of his clearest tests is what happens to the messenger. In a power-oriented culture, messengers are shot. In a rule-oriented one, they’re neglected. In what he called a generative culture, messengers are trained. The DORA research programme, which studies software delivery, found that a generative culture predicts better software delivery and organisational performance. So how you treat the person with bad news shows up in how well the team ships.
And Google’s Project Aristotle, which looked at more than 180 of its own teams, found psychological safety mattered more than anything else in what separated its best teams from the rest. It’s one company’s internal research, so I wouldn’t lean on it alone, but it points the same way as the other two.
Edmondson’s own advice for leaders, in her book The Fearless Organization, includes a step she calls responding productively: show appreciation, take the stigma out of failure, and still deal firmly with clear violations. That’s the reply, and the rest of this post is how to do it.
What to say in the first 10 seconds
You don’t need a script, but it really helps to have a first line ready, because the first line is the one your defensiveness writes if you don’t. These are the ones I use and teach:
- “Thank you for telling me.” Say it first, before anything else, and mean it. They took a risk.
- “What have you seen?” Ask for the detail before you give your view. You’ll often find the problem is bigger, or different, from what you assumed.
- “I can see why that’s frustrating.” Name the cost to them. It shows you heard the problem and not just the words.
- “What would you change?” The person closest to the problem usually has the best first idea, and asking treats them as part of the fix.
- “I’ll look at it and come back to you by Thursday.” A specific next step with a date. Then actually come back on Thursday, even if the answer is “not yet”.
And there are a few lines worth catching yourself before you say: “well, actually”, “that was a decision we made for good reasons”, “did you raise this at the time?”, and “everyone else seems fine with it”. Each one may be true, and each one tells the person they should have kept quiet.
The last step is the one most leaders skip. Close the loop. Go back and say what changed because they spoke up, even if it was small. People raise the second problem because of what happened to the first one.
1:1s, retros and incident reviews
The same reply works everywhere, but each setting has its own trap.
In 1:1s
Ask directly: “What’s one thing that’s slowing you down that I could fix?” Then write down the answer while they’re still talking. Bring the list to the next 1:1 and go through what you did about each item. After a few weeks, people start bringing things before you ask, because they’ve seen the list move.
In retros
Speak last, or not at all for the first round. If the most senior person opens with their view, everyone else adjusts to it. Watch the actions at the end as well. If they’re all “be more careful” or “communicate better”, the team is giving you safe answers. A retro that feels safe produces changes to process: a check added, a step removed, an owner named.
In incident reviews
Ask what happened before you ask who. A blameless review doesn’t mean nobody is accountable. It means you’re looking for the conditions that made the mistake easy, so you can change them. The best sign it’s working is when the person who made the change that broke production is the one who writes up the timeline, and they do it without being asked.
When the feedback is about you
This is the hard one, and it’s the one the team watches most closely. When someone tells you that you talk over people in planning, or that your last-minute changes are wrecking their sprints, every instinct says explain. Don’t, at least not yet.
Say thank you. Ask for an example, and listen to the whole of it. Then say what you’ll try: “I’ll hold my view until everyone else has spoken, and I’d like you to tell me if I slip.” That last part matters. You’ve turned a one-off complaint into permission to keep telling you.
If you think they’ve got it wrong, you can say so, but do it in a later conversation, once they’ve seen you take the first part seriously. As a fractional CTO I’m often the newest person in the room and the one with the least history, and I’ve found the quickest way to earn trust is to thank the first person who gives me feedback on something I did, in front of the team.
How to tell if your team feels safe
Engagement surveys tell you something, but they’re slow and people learn what to write. I’d watch what people do instead:
- Who disagrees with the most senior person in meetings. If the answer is nobody, that’s a warning, not alignment.
- How long bad news takes to reach you. In a team that feels safe, the gap between something going wrong and you hearing about it gets shorter.
- Whether problems come to you before they’re finished. “I think this estimate is slipping” on a Tuesday is a much better sign than “we missed it” on a Friday.
- Whether people raise things in public channels. Concerns that only ever surface in private 1:1s mean people are managing the risk of saying them out loud.
- Who leaves, and what they say on the way out. If the exit interview is the first time you hear about a problem, the reply to an earlier one taught them not to bother.
If you’re seeing more problems reported than before, that’s usually good news, for the same reason Edmondson’s hospital units looked worse on paper. The problems were always there. You’re just hearing about them now.
If you’d like a second pair of eyes on how your team handles bad news, or an outside read on what isn’t being said, book a call.
Frequently asked questions
How do you create psychological safety at work?
Mostly through how you reply when someone raises a problem. Thank them before you explain anything, ask what they've seen, say what you'll do and by when, then go back and tell them what changed. Do that every time, including when the problem is you, and people learn that speaking up is worth it.
What should a leader say when someone raises a problem?
Start with thank you for telling me, then ask a question rather than giving an answer, such as what have you seen or when did you first notice it. Leave your explanation until they've finished. Close by saying what happens next and when you'll come back to them.
How can you tell if your team feels psychologically safe?
Watch behaviour rather than survey scores. In a safe team people disagree with the most senior person in meetings, bad news reaches you quickly, the person who made a mistake is the first to say so after an incident, and retros produce changes to process rather than promises to be more careful.
Join the Drift Digest
One short email when a new post lands, plus new AI tells as models change and a quarterly nudge to re-measure your voice. No spam, unsubscribe anytime.
Something went wrong. Please try again, or email web@ctoondemand.co.uk.
