A team gets access to an AI assistant. Someone runs a workshop. There is a shared folder of prompts. Then Friday arrives, and the weekly report gets assembled the same way it always has: open the project board, copy the updates, check a document, rewrite the summary.

That is the moment worth studying. What would make AI useful enough for someone to change the way they finish that report?

MindStudio’s article on the AI adoption gap prompted this question. Our perspective at RideAlong is that adoption becomes easier to work on when you bring it down to the scale of one recurring task.

What the research actually says

In its 2026 study, IBM surveyed 2,000 CEOs and equivalent senior leaders across 33 geographies and 21 industries. Respondents reported that 25% of their workforce uses AI regularly, while 86% of the leaders believed employees had the skills to collaborate with AI. Source: IBM’s study announcement.

Those percentages describe different things: an estimate of workforce use and the share of CEOs expressing confidence in employee skills. They do not establish that 86% of employees have AI access, or that the difference measures a single population’s adoption gap.

The useful question for a team leader is still clear: where does confidence in AI fail to become a repeatable practice? The approach below is our recommendation for exploring that question, rather than a causal finding from the survey.

Choose one recurring task

Start with something a person can show you. Ask them to walk through the last time they completed it, with the actual inputs and finished output available. Look for the moment they have to gather, reconcile, or reformat information before they can move forward.

For the weekly report, that might mean collecting updates from three places and fitting them into a familiar template. Keep the first experiment small: use approved source material to produce a draft, with a person checking it before it goes anywhere.

Write down what a useful result needs to contain. A project update might require the current status, an owner, a deadline, and unresolved questions. Agree on what should happen when the source material is missing or contradictory.

The weekly report is an illustrative example, not a RideAlong customer result.

Guide the first attempt, in context

Plan the first attempt for the next time the task comes around. Have someone available to help select the right inputs, explain which data is permitted, and review the draft alongside the person doing the work.

A useful instruction is specific enough to act on:

Using these approved project updates, draft this week’s report in our template. Keep owners and dates tied to the sources. Flag missing information for me to check.

After generating the draft, compare it with the inputs. Did it turn a tentative date into a commitment? Did it leave out a blocker? Did it combine two projects that should stay separate? Each correction helps define what the next attempt needs.

Give the employee a clear way to decline the suggestion or return to their existing process. Their explanation is useful evidence about the task, the tool, or the guidance.

Measure the whole job

Record how long the task normally takes, then count the full assisted attempt: preparation, generation, checking, corrections, and handoff. A quick first draft can still create a slow review.

For a small pilot, keep a simple record after each attempt:

  • Completion: Did the finished output meet the agreed standard?
  • Effort: How much time went into the entire task, including corrections?
  • Friction: Where did the person need help or leave the workflow?
  • Repeat use: Did they choose to use the approach the next time?

Review a few repetitions before expanding. If the draft repeatedly needs extensive repair, narrow its job. It might be useful for organizing source notes even when it is not ready to write the final summary.

Make next week easier

At the end of the pilot, leave behind a short, usable recipe: the inputs, the approved tool, the instruction, the review checks, and the person to ask for help. Keep it where the task starts so the employee can find it again.

Then ask what still feels awkward. Copying material into a separate window, finding the right prompt, or remembering a review checklist can each become the reason an otherwise useful experiment never gets repeated.

This is the work RideAlong is built around: observing everyday activity, surfacing recurring workflows, and guiding people through using AI on the task in front of them. The team can then decide which suggestions are worth pursuing.

For your next AI adoption experiment, choose one person, one recurring task, and one clear definition of a useful result. Learn from what happens when that task comes around again.

Find your starting point.

Bring a messy week of work. We’ll talk through where RideAlong could help.

Talk to a founder

Sources & further reading

This is an original RideAlong perspective inspired by MindStudio’s article. Statistics above use IBM’s primary release; the workflow example and recommendations are ours.