A great training session can open a door. What builds real capability is walking through it, repeatedly, on work that matters. The two are very different things — and most AI rollouts invest heavily in the first while leaving the second to chance.
This isn't a flaw in your team. It's how humans learn any new skill. The people who become genuinely capable with AI aren't the ones who read the most about it or attended the best workshop. They're the ones who go through the same loop, over and over, on work that actually matters:
That loop is what you're creating conditions for as a manager. Your job isn't to push people to use the tool more. It's to make the trying feel low-stakes enough that people actually do it, and visible enough that they learn from each other as they go.
Capability comes from repetition on real work, not from understanding the tool. Your job is to create conditions where trying feels worth it.
One of the most useful things you can do as a manager is help your team figure out where AI will actually make a difference in their specific work — not in general, but for the tasks they do every week.
The best starting points share a few qualities:
Here are the four categories worth exploring with your team.
The most effective starting points are usually the ones your team finds slightly embarrassing to admit take so long — the routine tasks that eat time without adding much judgement. Those are the ones where AI saves real effort and where the quality bar is easy to meet.
Almost everyone's first few attempts with AI produce something that feels generic, slightly wrong, or just not quite right. That's not a sign the tool isn't for them — it's a completely normal part of learning any new skill. But if no one names it, people draw their own conclusion: the tool doesn't work for me.
Most of the time, the issue isn't the tool — it's the prompt. And that's good news, because prompting is a learnable skill.
A prompt is everything you give the AI before it responds — the instruction, context, examples, and constraints that shape what it produces. Think of it like a brief to a very capable but very literal assistant who knows nothing about you or your work unless you tell them.
- Give it a role and audience. "Help me write a team update for a non-technical audience" produces completely different results than "write a team update." The more the AI knows about who's reading it, the more it can tailor the output.
- Be specific about format. "Give me 5 bullet points, each under 15 words" is more useful than "make it concise." Concrete instructions produce concrete results.
- Add context that matters. What decision is this informing? What tone feels right? Any constraints? The more specific the brief, the more useful the output.
- Iterate rather than accept. A first output is a starting point. "Make the third point stronger", "rewrite in a more direct tone", "cut this by half" — refining through the conversation is the whole skill.
Two more moves worth trying once the basics feel comfortable: talking to AI instead of typing — you naturally give it more context at speaking pace than typing pace — and meta-prompting, where you paste in a rough attempt and ask the AI to rewrite it, questions first.
There are three moments where this still trips people up even with good prompts. Open each one to see what's going on and what helps.
Generic outputs come from generic prompts. When someone asks AI to "write a summary of this project," they'll get a competent but unmemorable summary. What they actually need to do is give the AI enough context to produce something specific — who's reading it, what they already know, what decision it's informing, what tone feels right.
Writing a good prompt is a skill, and it develops with practice. The first time feels awkward. The fifth time starts to feel intuitive.
This one shakes people — and understandably. AI can produce errors with great confidence, and if someone catches one (especially in front of others), trust drops fast. The risk is real, and the right response isn't to downplay it.
What helps is building the habit of checking outputs before using them — especially for facts, figures, or anything that will be shared externally. That's not a workaround; it's just how the tool works best. Checking AI output is part of the workflow, not evidence the tool is broken.
This is actually a sign of someone developing real judgment — they can tell the output isn't right, but they haven't yet learned the feel for when to refine versus when to reframe and retry. Most experienced users develop an instinct for this over time, but it's hard to transfer directly.
A rough heuristic that helps: if the structure and direction are right but the language is off, edit. If the structure itself is wrong, it's faster to start again with a more specific prompt.
The teams that build the most capability aren't the ones that scheduled the most AI time. They're the ones that wove it into existing rhythms — meetings, check-ins, shared docs. The tool became part of the rhythm rather than an addition to it.
Three moves that work well in practice:
Adoption that lasts is habit-driven. Mandates create compliance; habits create behaviour change. The difference is that habits are automatic — they run without you having to enforce them.
How habits form in workplace contexts: a habit needs a cue (something that triggers the behaviour), a routine (the behaviour itself), and a reward (something that makes the brain want to repeat it). For AI use, the cue might be the start of a specific task. The routine is using AI for that task. The reward is time saved, quality improved, or effort reduced.
The practical move: pick one task in your team's existing workflow where AI creates a clear, immediate benefit — and anchor the habit there first. Make it the obvious thing to reach for in that moment. Once it's automatic there, it spreads naturally.
Habits form when the reward is consistent. Starting with use cases where AI delivers a clear, reliable win gives the habit loop somewhere to take hold. That's why use case selection is a change management decision, not just a technology one.
Once a habit is anchored, some teams go further and get the AI to run its own loop — set an objective, a way to measure quality, and a limit, then let it iterate on its own instead of prompting it step by step.
Building the Habit
Your five answers, gathered from each section above. Edit them up there — this is a read-only copy for review, saving and printing.