Tom Hauser manages a regional field-service team: twelve technicians and three dispatchers who keep commercial HVAC systems running across four cities. When his company rolled out an AI assistant for drafting service reports and suggesting likely fault causes, Tom expected a productivity bump. What he got, in the first month, was near silence. Adoption sat at 18 percent. A few technicians tried the tool, hit a wrong suggestion, and quietly went back to doing everything by hand. One dispatcher used it for everything and started pasting AI-guessed diagnoses straight into customer-facing tickets without checking. The tool was fine. The culture around it was not. Tom realized his real job was not to deploy software. It was to shape how his team thinks, talks, and behaves around AI, so that good tools actually get used well.

Why culture decides everything

Culture determines actual behavior, and Tom learned that you can buy a tool but you cannot buy the conditions that make people use it well. Without a healthy AI culture, his team drifted toward two bad outcomes at once: some people avoided AI entirely and others used it recklessly, nobody surfaced problems because it felt risky to admit a mistake, and the one technician who figured out a clever prompt kept it to himself. With a healthy culture, the opposite happens. People use AI thoughtfully, balancing speed with care. They raise problems early because it is safe to. Knowledge spreads instead of being hoarded. And momentum builds because small wins are visible.

The most important thing Tom internalized: culture is shaped far more by day-to-day leadership than by any policy document. How he responds when a technician's AI-assisted report contains an error, what he praises in the morning huddle, how he makes decisions out loud - those signals teach his team what is really valued, regardless of what the official guidelines say.

The cultural elements that matter for a team

Tom focused on six elements, each of which he could influence directly without needing any executive mandate.

Psychological safety is the foundation. People have to feel safe taking a risk, asking a basic question, or admitting they do not understand something. When a technician told Tom "I honestly do not get how this thing decides what is wrong," Tom treated it as a reasonable question, not a sign of incompetence, because the alternative is people quietly faking competence and hiding errors.

Responsible experimentation means encouraging people to try the tool within clear guardrails. Tom's framing was "try it on a tenth of your reports this week and tell me what happens," not "do not touch it without sign-off." Experiments had a clear question and an honest read of the result, and a failed experiment was a lesson, not a black mark.

Knowledge sharing turns one person's discovery into the whole team's gain. When a dispatcher found that pasting the equipment model number into the prompt produced far better fault suggestions, Tom made sure that tip reached everyone instead of staying a private edge.

Celebrating responsible innovation shapes behavior because what gets praised gets repeated. Tom deliberately celebrated not just speed but speed with judgment, and he publicly recognized the technician who decided an AI suggestion was wrong and overrode it. Rejecting a bad AI output is a win, not a failure.

Discouraging recklessness means being clear, without shaming anyone, about what does not fit: blindly trusting output, automating something with no check, ignoring an obvious accuracy problem, or letting AI text reach a customer unread. "This is how we work here" is a kindness, not a reprimand.

A growth mindset keeps people from sorting themselves into "AI people" and "not AI people." Tom's line, repeated often, was "your expertise is exactly what makes the AI useful; it changes how you work, not whether you are needed." He celebrated learning - taking the training, trying a tool - as much as immediate results, and he openly admitted he was learning too.

Worked example: running the change with ADKAR

Tom needed a structure, so he used ADKAR, a change model that says any individual moves through five steps in order: Awareness, Desire, Knowledge, Ability, and Reinforcement. A change stalls at whatever step is weakest, so Tom rated his team on each, then targeted the gap. He scored each step as the rough percentage of his fifteen-person team who had reached it.

• Awareness (people know the tool exists and why): 90 percent. The rollout email and a team meeting had covered this well.
• Desire (people actually want to use it): 35 percent. This was the real gap. Several technicians saw it as extra clicks that threatened the craft they were proud of.
• Knowledge (people know how to use it): 50 percent. Some had watched the training; others had not.
• Ability (people can use it well in real work): 30 percent.
• Reinforcement (the new behavior sticks): 15 percent.

ADKAR told Tom not to push harder on awareness, which was already fine, but to attack Desire first, because no amount of training fixes a team that does not want the change. So he spent two weeks on Desire: he reframed the tool as a way to cut the tedious report-writing that technicians hated, not as a replacement for their diagnostic skill, and he had two respected senior technicians who liked the tool share concrete time savings in the huddle. Only once Desire climbed did he invest in Knowledge with a hands-on 30-minute session, then Ability through paired practice on real jobs, then Reinforcement by making "what did the AI help with this week, and where did you correctly override it?" a standing huddle question.

The sequencing mattered. A month earlier Tom had thrown more training at the problem and adoption barely moved, because the blocker was Desire, not Knowledge. By naming the weakest step and fixing it first, he stopped wasting effort on steps the team had already passed.

A simple culture-norms scorecard

Culture can feel too vague to manage, so Tom built a one-page scorecard he revisited monthly. He rated six norms from 1 to 5 and watched the trend, not the absolute number.

• Psychological safety: can people tell me about an AI mistake and how I responded? Started at 2, reached 4 by month three.
• Experimentation: are people trying the tool within guardrails? Started at 2, reached 4.
• Knowledge sharing: do discoveries reach the whole team? Started at 1, reached 3.
• Celebrating the right things: do I praise judgment, not just speed? Started at 2, reached 4.
• Discouraging recklessness: is unread AI text reaching customers? Started at 2, reached 5 after he set a clear "every customer-facing line gets human review" norm.
• Growth mindset: are people saying "I can learn this" rather than "that is not me"? Started at 2, reached 4.

The scorecard turned a fuzzy feeling into something Tom could act on. The lowest-scoring norm each month told him where to put his attention next, and the rising trend gave him and his team visible proof that the work was paying off.

The everyday practices that built it

None of this came from a memo. It came from a handful of repeated behaviors. In hiring and onboarding, Tom screened new technicians for learning agility and curiosity rather than prior AI skill, and every new hire learned "how we use AI here" in their first week, paired with a peer who modeled it. For training, he ran a short monthly session and kept it non-threatening, using real service jobs as examples rather than abstract demos. In decisions and the daily huddle, he asked critical questions out loud - "have we double-checked this diagnosis the AI suggested?" - and welcomed healthy skepticism rather than treating it as resistance.

Recognition did heavy lifting. Tom recognized the technician whose careful skepticism caught a wrong AI fault code, the dispatcher who shared a better prompt, and the person who spent an hour learning a feature that turned out not to help, because that hour was still good judgment. He used stories deliberately: "Remember when the tool suggested a compressor fault and Dana checked it and found a sensor issue instead? That is exactly the thinking we want." And he modeled it himself, openly admitting when he did not understand something and saying "I am learning this alongside you," because a manager who pretends to have all the answers teaches everyone else to pretend too.

Governance, at his level, was light but real. When someone escalated a concern about a wrong AI suggestion, Tom responded with curiosity, not defensiveness, and treated the escalation as evidence the team was catching problems early. When something went wrong, he ran a blameless five-minute review focused on "what do we change?" rather than "whose fault was it?" That single habit, more than any rule, is why his people started surfacing problems instead of hiding them.

Putting culture into your systems, not just your tone

Tone sets the mood, but systems make culture survive a busy quarter. Tom built his into four places where behavior actually gets decided.

Hiring and onboarding come first, because culture starts with who joins and how they are introduced. Tom hires for learning agility, curiosity, and comfort with ambiguity rather than deep AI expertise, which he can teach. In interviews he asks two questions on purpose: how do you approach a new technology, and tell me about a time you raised a concern about something. He also keeps his interview panels mixed rather than stacking them with the most technical people, because a panel of enthusiasts selects for enthusiasm and misses the careful operator who catches what a tool gets wrong. A new hire's first week includes an explicit conversation about how the team thinks about AI, they are paired with a peer mentor who embodies it, and their first project is deliberately low stakes so they learn the culture in practice rather than in a slide deck.

Training comes next, and it works better in tiers than as a single event. Everyone needs a baseline understanding of AI fundamentals. Beyond that, training is role-specific, because a dispatcher, a technician, and a manager need different things from the same tool. It has to be ongoing rather than one-off: a monthly tip, a quarterly update on new capabilities, an annual refresher on what has changed. And there should be an optional advanced track for the people who want to go deeper, so the naturally curious have somewhere to go without the whole team being dragged along.

Decision-making is the third system, because how you decide signals what matters. Tom deliberately includes voices beyond the tech enthusiasts, asks the critical questions out loud ("have we thought about fairness here, and what could go wrong?"), and treats healthy skepticism as a contribution rather than an obstacle. He also documents his reasoning in a couple of lines whenever a real decision is made, because a written "here is why we chose this" lets the team learn from the decision instead of only seeing its outcome.

Knowledge sharing is the fourth, and it needs mechanisms rather than good intentions. That means a regular slot where people share what they have learned about AI, some written record of lessons that captures what worked, what did not, and why, cross-team mentoring where the person who is genuinely good at prompting helps others get there, and space for informal communities of practice to form around people learning the same thing at the same time.

Three situations you will probably recognize

Culture work looks different depending on where your team is starting. Three patterns come up repeatedly, and each calls for a different opening move.

The team that says "we do not do AI." Suppose your group has been openly skeptical, and the prevailing attitude is that they do not need this. The instinct is to argue. The better move is to acknowledge the culture you have: their skepticism has real value, because being thoughtful rather than jumping on trends is how good work gets protected. From there, reframe the question. It is not "should we use AI?" but "where can AI make us better while we keep our standards?" Then start small and safe by picking one or two low-risk areas, internal productivity work rather than anything customer-facing, and asking for two weeks and honest feedback about what works and what does not. Celebrate the responsible skepticism you get back, out loud: someone raising a fairness concern is exactly the thinking you want, and deciding a tool is not right for you is a smart decision rather than a failure. Support the advocates without forcing anyone, and let the early adopters prove the value. The result you are aiming for is not the disappearance of skepticism; it is a team that keeps its skepticism and opens up to thoughtful use.

The team that is willing but intimidated. Here people want to use AI but feel the technology is above them. Start with fundamentals in a non-threatening format, a monthly session that answers what AI is, when it is useful, and roughly how it works, using examples drawn from the work they actually do rather than abstract demos. Normalize not knowing: nobody has this figured out yet, and you are all learning together. Build peer learning by pairing less-confident people with early adopters and naming who is good at what, and open a shared channel where questions and tips can be asked without ceremony. Share successes in a way that includes the process, not just the outcome, because "they ran five experiments, most failed, and the one that worked came out of what the failures taught them" is far more useful than a headline number. Then lower the cost of trying: invite people to use a tool on 5 percent of their work as a safe place to fail, tell them plainly that three hours spent learning something that did not pan out was a good use of time, and allocate real time for it, on the order of a tenth of the week for learning and experimentation. Celebrate the learning itself, so that taking the advanced training or helping a colleague get started counts as a contribution.

The team growing fast. When headcount rises quickly, the risk is that the values you built get diluted by people who never absorbed them. Be intentional in hiring, screening for learning agility, curiosity, and responsibility rather than only for AI skill. Let onboarding carry the culture, with the first-week conversation about how you think about AI, a peer mentor who lives it, and a low-stakes first project. Reinforce it regularly through stories at the monthly all-hands and a quarterly culture check that asks plainly what is working and what is drifting. Expect governance to grow with scale rather than shrink; as the team gets larger and the risk gets bigger, you need more explicit controls and escalation paths, made visible and applied consistently. And keep celebrating the culture in public while naming drift when you see it, including the uncomfortable version: "that decision moved too fast without thinking through who it affects." Done well, new people pick the culture up quickly, because it is embedded in systems and behavior rather than in a founding story nobody was around for.

The traps Tom watched for

Tom kept a short list of cultural anti-patterns on his desk. A blame-based reaction to mistakes makes people hide problems, so the same errors recur; he chose blameless learning instead. A speed-only culture makes people cut corners and trust AI blindly; he celebrated speed with thoughtfulness. Stating values without enforcing them through recognition and accountability makes the values meaningless, so he tied them to what he praised and corrected. Treating some people as "naturally AI" and others as obstacles loses exactly the experienced judgment that catches what a tool misses; his veteran technicians spotted risks the tool never would. And treating culture as a nice-to-have while focusing only on the tool guarantees the tool gets deployed and never adopted, which was precisely his month-one problem.

He also runs quick gut checks. Can a team member describe a mistake they made with AI and how he responded? If that story does not exist, safety is weak. When someone learns something, does it reach the team, or stay private? Sharing signals a strong culture; hoarding signals a weak one. Can he tell three real stories about his team and AI - a success, a failure, a learning? If not, he is not building culture on purpose. And if he asked his newest hire "what is our approach to AI here?", could they answer? If not, onboarding did not embed it.

Fairness, inclusion, and the values-alignment test

The check Tom finds hardest is the one about his own consistency. Name the values you claim to hold around AI, whether that is responsibility, learning, or fairness. Then try to point at specific hiring, recognition, and accountability decisions that reinforce each one. If you cannot find the evidence, your culture is drifting, because a value that never changes a decision is a slogan. The gap between stated culture and actual culture is not usually a lie; it is neglect, and it closes only when the value shows up in who gets hired, who gets praised, and what gets corrected.

Three responsible-AI commitments belong inside the culture rather than beside it. Fairness has to be lived rather than stated, which means asking whether your hiring, recognition, and governance decisions genuinely reinforce it. Inclusivity in decisions means the people shaping how AI gets used include more than the technical enthusiasts, because the operations veteran and the newest dispatcher see risks the enthusiasts do not. And psychological safety has to extend specifically to escalation: people report a problem with an AI output only when they trust that raising it will not be held against them. Those three, held together, are what make the difference between a team that catches a bad AI output before a customer sees it and a team that discovers it afterward.

Where it landed

Six months in, adoption had climbed from 18 percent to 74 percent, and just as important, the reckless use had stopped, because the norm that every customer-facing line gets a human read was now simply how the team worked. Report-writing time per job dropped by roughly a third, freeing technicians for the diagnostic work they actually valued. The change was never really about the AI assistant. It was about a service manager deciding that culture is something you lead, day by day, through what you ask, what you praise, and how you respond when something goes wrong.

Terms worth keeping straight

• Psychological safety. An environment where people can take interpersonal risks, ask questions, and admit mistakes without fear of negative consequences.
• Responsible experimentation. Structured learning by trying new approaches inside clear guardrails, with a stated question and an honest reading of the result.
• Knowledge sharing. Information and learning distributed across the team rather than hoarded by whoever discovered it.
• Growth mindset. The belief that abilities are developed through effort and practice rather than being fixed traits you either have or lack.
• Blameless culture. Treating incidents as learning opportunities rather than occasions to assign fault.
• Culture drift. Values that are stated but not reinforced through action, so that the actual culture quietly diverges from the stated one.

Practice and Reflection

Culture work only becomes real when you apply it to the team you actually lead. Work through these five exercises for your own group.

Assess where you are. Write down your team's stated values around AI, then write down how people actually behave around it. Look hard at the gap between the two, and be specific about what is driving that gap: is it what you praise, what you tolerate, what leadership above you signals, or simply that nobody has ever said out loud what the expectation is?

Rate your culture indicators. Take the six elements from this lesson, psychological safety, responsible experimentation, knowledge sharing, celebrating responsible innovation, discouraging recklessness, and growth mindset, and rate your team on each from 1 to 5. Note where you are strong, where you are weak, and, for each weak area, what a 4 would actually look like in behavior rather than in words.

Design the practices. For the weakest areas, design concrete practices rather than intentions. How exactly will you celebrate responsible innovation, and in what forum? How will knowledge get shared, and by whom? What will you do differently to reinforce psychological safety? And what specific behavior of your own will model the culture you are asking for?

Write your hiring and onboarding flow. Name the traits and values you will screen for, how a new hire will learn the culture in their first week, who their culture mentor will be, and what first project will let them practice it at low stakes.

Collect your culture stories. Identify four real stories you can tell: a positive one where someone did it right, a learning one where something went wrong and the team improved, a challenge one where someone raised a concern that slowed things down and was correct, and a transformation one following a person's journey from wary to capable. If you cannot fill all four slots, that absence is itself a finding.

Then spend two minutes on the smaller reflection. Think back over the past week and find one decision, conversation, or reaction of yours where the ideas here would have changed your approach. What would you have done differently, and what would have changed as a result? That link between concept and your own behavior is where culture work actually starts.

Related Lessons

Culture sits underneath most of the other leadership work in this level, so several lessons connect closely to it.

• Leading AI Transformation covers the broader change effort that culture has to carry. Culture change is consistently the hardest part of any transformation, and it is where most stalled rollouts are actually stuck.
• Workforce Development and Reskilling supplies the development side of the culture you are describing. Training and career paths are how a growth mindset stops being a slogan and becomes something people can act on.
• Ethical Leadership in AI Adoption goes deeper on leadership modeling, which is the single largest driver of what your team believes is acceptable behavior with AI.

Key Takeaways

• Culture decides whether good tools get used well. You can buy an AI tool, but not the conditions that make people use it thoughtfully. Without culture, a team splits into avoidance and recklessness at once.
• Day-to-day leadership shapes culture more than policy. How you respond to a mistake, what you praise in the huddle, and how you decide out loud teach your team what is really valued.
• Psychological safety is the foundation. When admitting confusion or a mistake is safe, people surface problems early; when it is not, they hide them and the same errors recur.
• Use ADKAR to find the real blocker. A change stalls at its weakest step. Rate Awareness, Desire, Knowledge, Ability, and Reinforcement, then fix the gap; throwing training at a Desire problem wastes effort.
• Make culture measurable with a simple scorecard. Rate a handful of norms monthly and watch the trend. The lowest score tells you where to focus next, and the rising line proves the work is paying off.
• Celebrate the right things. Praise judgment and responsible skepticism, not just speed. Recognizing the person who correctly overrode a wrong AI suggestion teaches the behavior you want.
• Respond to escalations with curiosity and run blameless reviews. That single habit is what makes people report problems instead of hiding them.
• Everyone's expertise still matters. Experienced people catch risks a tool misses. A growth mindset, not an "AI people versus not" split, keeps that judgment in the room.