Psychological safety is the real AI adoption bottleneck (not another Copilot license)

A manager asks who has changed a workflow and the room goes quiet. The team already has the licence and a channel full of impressive screenshots. People have learned that a rough draft, a wrong number or a killed experiment can cost them. They wait until the work looks finished. Buying another seat on the same product leaves that calculation where it was.

Amy Edmondson calls this a shared belief that the team is safe for interpersonal risk. Google's Project Aristotle later found the same pattern in its own teams. People speak up, they admit a miss, and they try a half-formed idea when they believe the team will spare them the punishment. Work with a model is full of those risks. A quiet room has often done the maths.

Why the room gets quieter

Using a model on real work asks a person to do things that can cost them socially. They show a draft that might look naive. They admit they do not know whether the answer is right. They challenge a colleague's output, sometimes a senior colleague's, in a meeting where people are praising speed.

If the last person who showed unfinished work got a joke at their expense, they will hide the next draft. If the last person who said the number looks invented got labelled negative, they will paste the number. If the only stories in the channel are flawless wins, everyone else keeps the experiment off the record. The licence gives them a private place to hide it. You get shadow use, and a public story that adoption is low.

When we work with teams we keep the standard in the same breath as the permission. Give people safety and drop the standard, and they stay pleasant while the work gets worse. Keep the standard and punish a rough draft, and they go quiet. You want a room where people can say the output is weak, and where you still expect them to make it good enough to send.

The first month

A town hall line about it being safe to experiment never sets the tone. The moves that do are local, and they repeat. Show your own rough draft, including one place the model was wrong, before you ask the team to show theirs. In the same meeting where you ask what it sped up, ask what it missed. When someone catches an error, ask about the check. A speech about carelessness teaches the room to hide the next one.

Keep a visible example of a stopped workflow, so ending a bad use earns the same respect as launching one. Separate the experiment from the performance review for a defined period, and say the dates out loud. If a clumsy first month can show up as a lack of capability at review time, rational people will wait until they are fluent in private. You will conclude the team is slow, and you will buy more training. They were waiting until a clumsy month could stay out of the review.

What a learning team can run

You cannot install this with a module called trust. You can design the session so people practise the risk with the colleagues they will see on Monday. Put the intact team in the room with their manager. Use their work. Every person brings one output they would still hold back from a customer. Spend more time on the critique than on the prompt. End with a norm the team writes. How we show unfinished work. How we challenge a number. How we admit a miss. A culture deck will sit unread on the day it matters.

The psychological safety workshop is where we practise that habit. It sits beside AI training. If leaders send the team and stay outside, you have wasted the day. The person with the most status has to do the first risky thing, where everyone can see the response. What that day should change is in what a psychological safety workshop should change.

What to count

Logins measure access. Ask whether people will put a weak draft on the table, whether a junior can challenge a senior's assisted slide, and whether a killed experiment gets airtime. A manager who is looking can see those in a fortnight.

Ask four questions, anonymous, before an AI push and again six weeks later. Can I show unfinished work here? Can I say I do not know if this answer is right? Will anyone blame me for a good faith miss? Do we talk about what we stopped? If the scores are low, more seats will sit unused. Fix the room, then teach the workflow.

Keep the standard in the same sentence as the permission. People should be able to show unfinished work and say an output is weak, and you still expect them to make it good enough to send. Punish a thoughtful attempt and the next one happens in a personal account. Drop the standard and the team gets faster at sending work it has skipped the check on. The argument about the draft has to happen where the manager can hear it. The customer still gets a checked answer.