Forget the AI Strategy Offsite. Autopsy One Task Instead.

Most AI strategy conversations happen in a room with no computer open. Someone presents a roadmap, three quarters get mentioned, everyone nods, and by Monday morning nothing about anyone's actual work has changed. The room agreed on a direction. Nobody tested anything.

The fix is not a better roadmap. It is picking one real task and taking it apart in front of the people who do it.

This is a workshop we run at MAKIA Labs when a team already believes in AI in theory but has never actually watched it work, or fail, on their own material. It borrows the bones of a design thinking exercise: decompose, test, decide, together. The output is not a slide. It is a written sequence a team can start using the same week.

Objective

Break one real, recurring task into its individual steps, and decide for each step whether it stays fully human, gets augmented by AI, or can be automated. The team leaves with a tested, adoptable micro workflow, not an abstract policy.

Participants

Four to eight people who actually do the task, or are directly affected by it. A single team works best. Mixing departments works too, as long as everyone in the room touches the same process somewhere.

Duration

Ninety minutes. Do not stretch it. The time pressure is what stops the conversation from drifting into theory.

Materials

A whiteboard or shared digital board, sticky notes in three colors, a laptop with an AI chat tool already open and logged in, a timer visible to the room, and real work material the team can use during the live test. Anonymize anything sensitive beforehand.

Run sheet

Minutes 0 to 5. Frame it. Explain the rule: one real task, not a hypothetical, not how we do sales in general. Something a person in the room could describe in one sentence.

Minutes 5 to 15. Select the task. The group votes on one recurring task everyone touches. Pick something mildly annoying rather than mission critical. Lower stakes mean people experiment honestly instead of defending existing habits. Write the task as a single sentence at the top of the board.

Minutes 15 to 30. Break it down. Individually first, then as a group, decompose the task into six to twelve concrete steps. One sticky note per step, arranged left to right in the order they actually happen, not the order they are supposed to happen.

Minutes 30 to 50. Sort each step. For every step, the group decides together: keep human, because it needs judgment, a relationship, or accountability nobody wants to hand off. Augment with AI, meaning a person still does the step but AI drafts, checks, or suggests. Or automate, meaning no human involvement is needed most of the time. Where the group disagrees, use dot voting rather than letting the most senior voice decide alone. That disagreement is the most useful data point of the whole session.

Minutes 50 to 70. Live test. Pick two or three steps marked augment with AI. Actually do them, in the room, on real material, typing the real prompt and reading the real output together. Not a demo prepared in advance. The point is watching where it works and where it quietly does not.

Minutes 70 to 80. Build the micro workflow. Write the resulting sequence as a short numbered list: who does what, in what order, with which tool. Specific enough that someone could follow it on Monday without asking a single clarifying question.

Minutes 80 to 90. Debrief.

Debrief questions

Which step surprised you, one you assumed was AI's job that clearly needed a human, or the reverse.

Where did the live test disappoint you, and what would the output need to look like before you would trust it without checking.

Who owns keeping this micro workflow alive after today.

What would it take to run this same exercise on a second task next month.

Common failure modes

Picking a task too large or too political, like how we handle customer complaints instead of one specific complaint type. The facilitator should veto anything that cannot be written as a single sentence under fifteen words.

Skipping the live test because it feels slower than just discussing. The exercise has no value without it. Watching real output land, or fail to land, is the only part people actually remember by Friday.

Letting one person, usually the most senior, decide every column alone. The exercise exists to expose where perceptions of AI's usefulness differ across a team, not to confirm one person's existing opinion.

Ending without a named owner for the resulting workflow. Without minute 80's ownership question, the sticky notes get photographed for posterity and nobody opens the photo again.

This connects to the Knowledge and Impact dimensions of the MAKIA framework more directly than most exercises we run. Knowledge, because the live test shows a team exactly what the tool can and cannot do on their own material, not a vendor's demo material. Impact, because the group is deciding, task by task, whether a change is actually worth making, not whether AI is impressive in the abstract.

Try this week

Pick one recurring task from your own week, describe it in under fifteen words, and privately sort its steps into the three columns before you run this with anyone else. Notice which step makes you hesitate. That is the step worth testing live first.