For years, learning a new tool at work meant a course. Someone booked a session, a link arrived, and the plan was to sit through the modules and then, eventually, start using what we had learned. The assumption underneath was simple: knowledge comes first, practice comes second.
With AI, that order seems to be flipping. In a study of 767 knowledge workers, people who knew at least one person using AI were about three times more likely to have used it themselves in the past week than people who knew no one. Nearly nine in ten participants did know someone. The strongest signal for starting was not a course, a licence or a policy. It was being close to somebody already doing it.
That is a time issue, and a bigger one than it looks. The gap between "I should try this" and "I just finished something real with it" is where hours quietly disappear. A month of good intentions is a month of saved time that never started. Ten minutes beside a colleague can close that gap on an ordinary Tuesday afternoon.
------------- Context -------------
Most teams still treat AI adoption as a training problem. Budget goes to courses, prompt libraries and lunch-and-learn sessions. The thinking is reasonable: if we can get everyone through the material, the use will follow. Completion rates become the scorecard, and a high completion rate feels like progress.
But courses carry a hidden timing problem. They ask for hours up front and return the payoff later, if at all. Picture a three-hour course spread over two weeks, followed by a gap while the learner hunts for a real task to apply it to. At the end, that person has spent three hours and saved none. Plenty of us have been that learner, and plenty of us never opened the second half.
This is where it helps to name a different measure. Call it time-to-first-use: the number of days and minutes between deciding to try something and finishing a real piece of work with it. Courses tend to stretch that distance. Watching a peer shrinks it, because the learner sees the tool working on a real task before they have to figure out the task themselves.
The reason is that a colleague answers several questions at once. Which task is worth trying? How do we phrase it? What does a good first result look like, and what does a weak one look like? When do we stop and take over? A course answers these in general. A colleague answers them on our actual work.
That matters because the time we reclaim from AI depends on how quickly the first real use happens. If we measure the distance in days instead of completion rates, we start asking a more useful question: who on our team could show someone else this week?
------------- A Course Answers the Whole Topic, a Colleague Answers Our Tuesday -------------
One of the quietest costs in learning anything new is translation. A course teaches the idea in a neutral example. We then have to carry it over to our own files, our own clients and our own awkward formatting. That translation work is invisible, but it takes real minutes.
Consider an operations coordinator who finishes a module on summarising documents. The example was a tidy two-page report. Her real weekly supplier update is eleven messy emails and a spreadsheet. She spends around 40 minutes working out how the lesson applies, gives up halfway, and goes back to doing it by hand.
Now picture her sitting beside a colleague who already does this. In ten minutes she watches him paste in the same kind of emails, ask for the three things that changed, and check the result against the spreadsheet. She does not need the theory. She needs the move, shown on material that looks like hers.
This is a direct time gain. The first use drops from roughly 40 minutes of translation to 10 minutes of watching plus a short attempt of her own. After that, the weekly update that took 50 minutes by hand takes closer to 20. Across a month, that is two hours back, and the first 30 of those minutes arrived in the first week rather than the sixth.
------------- Seeing a Messy First Result Teaches More Than a Perfect Demo -------------
The best demos in a course are rehearsed. The prompt is polished, the output is clean, and the whole thing takes ninety seconds. That is useful for showing what is possible, but it hides the part that actually costs us time: the second attempt.
When we watch a colleague work live, we see the real shape of the task. The first answer is a bit off. They tighten the instruction, discard a paragraph, ask again, and fix a number by hand. None of that looks impressive. All of it is the most valuable thing we could learn, because it tells us what normal looks like.
Without that picture, a new user often assumes a weak first result means they did something wrong. They stop, or they spend 20 minutes rewording the same request in silence. Seeing someone else treat a rough first draft as an ordinary starting point removes that doubt in about thirty seconds.
That has a clear time effect. Restart time falls, because we know a second pass is expected rather than a failure. Review time becomes more consistent, because we have watched how a person checks the work before trusting it. A new habit that might have taken three or four frustrating sessions to form can arrive in one.
------------- The Person Next to Us Is a Channel We Rarely Use -------------
If nearly nine in ten of us already know someone using AI, then the learning material is sitting in the room. Almost every team has one or two people who have quietly built a workflow that saves them time. Most of them have never been asked to show it, and most would not think to offer.
The usual reason is that sharing feels like a big commitment. A "training session" sounds like slides and preparation. A ten-minute screen share does not. When the ask is small, it is easy to say yes, and the cost to the person showing is close to zero because they are only doing their normal work out loud.
Run the numbers for a team of eight. One person shows a single workflow, and seven colleagues watch for ten minutes each. That is 70 minutes of combined attention. If even four of those seven adopt the workflow and each saves 30 minutes a week, the team recovers two hours a week. The 70 minutes is repaid in the first week, and the saving repeats.
The pattern also compounds. Each person who learns by watching becomes the next person who can show. The distance between the first user and the last shrinks every round, which is the opposite of a course rollout, where the last cohort often learns months after the first.
------------- Borrowed Use Becomes Ours Only When We Run It Again -------------
There is one limit worth respecting. Watching someone else is a fast way to start, but it is not the same as owning the workflow. A borrowed prompt works well on the colleague's material and sometimes only half-works on ours. If we never run it ourselves, the lesson fades within days.
The fix is short and specific. Within a day of watching, we run the same task on our own work, even if the result is imperfect. That second step turns a demo into a habit. It also surfaces the small adjustments that make the workflow fit, such as a different tone, a different file type or a different review step.
Think of a consultant who watches a peer summarise client calls, then tries it on her own call notes the next morning. The first attempt misses two action items. She adds a line asking for owners and deadlines, and the second attempt works. That adjustment takes about 15 minutes once, and every call afterwards saves her around 20 minutes of note cleanup.
This is where the time gain locks in. The 10 minutes of watching saves the translation cost, the 15 minutes of adjusting saves the fit cost, and the repetition turns both into a saving that comes back every week. Skip the second step and the whole thing stays a nice demo.
------------- Practical Moves -------------
First, pick one person on our team who already uses AI on a real task and book ten minutes of their screen this week, because a booked ten minutes beats an open-ended intention to learn.
Second, ask them to show the task live, including the rough first result, the correction and the point where they stop, so we see review time as well as speed.
Third, run the same task on our own material within 24 hours, since a borrowed workflow only saves time once it has been used on our work.
Fourth, note how long the first real use took from the moment we decided to try, and compare it with the last time we tried to learn a tool from a course.
Fifth, pass it on by showing one other colleague within the week, so every person who learns shortens the next person's time-to-first-use.
------------- Reflection -------------
The finding that people who know an AI user are about three times likelier to use AI themselves is a signal that adoption is less about information and more about proximity. We have plenty of information. What many of us lack is a short, low-pressure moment of seeing the thing work on something that looks like our job.
That is why this ties so directly to time. Courses ask for hours before they return anything. A peer demonstration asks for ten minutes and returns a real result the same day. When we shorten the distance between deciding to try and finishing something real, the weekly savings start weeks earlier, and they keep adding up from there.
In the end, that is the kind of AI shift worth paying attention to. Not a bigger library of lessons, but a smaller gap between the lesson and the first useful result. The people who already know how are a faster route than we tend to give them credit for.
Who on your team or in your circle is already using AI in a way you have never actually watched?
What is one task of yours that a ten-minute screen share could have saved you an hour on?
If you showed one workflow to a colleague this week, which would give them the fastest first win?