Before your team argues about AI, sort the tasks

Marco van Hout · 4 min read

Teams that ask "should we use AI?" end up with guidelines nobody follows. A 90-minute session that sorts recurring tasks into what AI drafts, what it assists and what stays with people, and ends with one measured two-week test.

Most teams start the AI conversation in the wrong place. Someone asks "should we use AI for this?", half the room says yes, the other half worries, and the meeting ends with a promise to write guidelines. Two months later a few people use it daily, most never touched it, and the rest clean up after both.

The question is too big to answer in one go. "Which of our tasks, and how would we know it worked?" is a question a team can answer in 90 minutes.

Saved time is not the same as better work

Two surveys from October show why the vague version fails. Hubstaff asked 1,068 US office workers about their workweek. 79% use AI tools at work, and 76% of those who report savings gain at least an hour a week. But only 26% use that time to take on more work, and 46% say expectations of their workload went up after AI arrived. 45% of AI users spend an hour or more a week fixing colleagues' AI-generated work.

Gallup, surveying more than 15,000 US employees, found that 28% use AI at least weekly and 54% have never used it in their role. Regular users report far more benefit than occasional ones: 79% of them say AI makes them faster, against 38% of irregular users. And 52% of all employees say they have less say over new technology at work than they would like.

Put together: the gains go to the people who already use AI, the rework lands on everyone, and most people feel the decision was made somewhere else. A session fixes the last point and can make the other two visible.

A 90-minute session: sort the tasks, pick one test

For a team of 5 to 15 who do related work. You need a wall or a shared board and sticky notes in three colours.

  1. 0:00-0:10. The question. The lead says what the session will decide: one task the team will test with AI for two weeks, and how the result will be checked. Not a policy, not a tool choice.
  2. 0:10-0:30. Task inventory. Everyone writes, alone and in silence, the recurring tasks in their week that take more than 30 minutes. One task per note. Cluster duplicates on the wall. Aim for 15 to 30 tasks.
  3. 0:30-0:50. Sort into three columns. "AI drafts, a person decides", "AI assists, a person does the work", "stays with people". Work in pairs, one column per round. Any task the room disagrees on gets a red dot and stays in the middle.
  4. 0:50-1:05. Talk about the red dots only. For each one, ask: what would go wrong if AI got this slightly wrong, and who would notice? Move it or leave it in the middle. Middle tasks are not tested this round.
  5. 1:05-1:20. Pick one test. From the first column, choose one task that happens at least weekly and where a mistake is cheap and visible. Write down three things: who runs the test, what "worked" means in numbers (minutes saved, rework minutes, error count), and who checks the output before it leaves the team.
  6. 1:20-1:30. Close. Book a 20-minute review exactly two weeks out. Put the wall in a photo and share it the same day.

Rules you can check two weeks later

  • One task per test. A second task waits for the review.
  • Measure both sides: time saved by the person using AI, and time spent by others checking or fixing the output. If rework is more than half the saving, stop or change the task.
  • Every AI-drafted output has a named person who signs it off before it goes outside the team.
  • At least half the team takes part in the test at least once, not only the regular users.
  • At the review, decide one of three: keep, change, stop. Then run the next test from the same wall.

When the room splits

Some people will want to talk about jobs, ethics or data before any test. Write those concerns on a separate sheet at the start and give them a slot at the review. Tasks that touch personal data, client confidentiality or legal commitments go straight to "stays with people" until someone with authority says otherwise. That keeps the first test small enough to run.

Where to start

Open Session Studio with your team size and the time you have, and ask for a session where the team sorts its recurring tasks into what AI drafts, what AI assists and what stays with people, ending with one two-week test and how it will be measured.

Sources

Design your own session

METODIC turns ideas like these into a complete session agenda with activities, timing, and materials — for workshops, meetings, offsites, and team sessions.

Try METODIC free