Learn
The AI In-Basket takes the in-basket, a proven method, and adds a new colleague: AI.
In the mid-1950s, Norman Frederiksen and his colleagues at Educational Testing Service, in the United States, designed a test to train and assess US Air Force officers: the in-basket exercise. They published it in 1957.
The idea is simple. The person plays a manager who has just started a new job. They find a pile of documents: memos, letters, messages. Time is short. For each document, they decide: act, delegate, put it off or let it go. Then they explain their choices.
In 1956, AT&T used a version of the exercise in its Management Progress Study, the first known use of assessment centres in business. The exercise became a classic of management selection.
In the classic in-basket, one of the choices was already "delegate". With AI, there’s a new colleague to delegate to. It’s fast, always available and often good. It can also be wrong with total confidence.
The AI In-Basket keeps the spirit of the exercise: a pile of real tasks, little time, choices to justify. But instead of assessing one person, it gets a team talking. There’s no score. There’s a conversation.
Each bin is a way of working with AI. The lines between them aren’t always sharp. That’s on purpose.
The task stays human. AI can sometimes prepare a small piece of it, but the task itself needs a person.
It’s often the case when
Example: Calling back an employee whose father just died.
Watch out: "I keep it" doesn’t mean "I do it all alone". AI can remind you of a rule or review your draft.
AI writes a first draft. You correct it, fill in the gaps and decide. It’s often the bin that pays off most.
It’s often the case when
Example: Comparing two 38-page quotes and pointing out the differences.
Watch out: Keep time to check. A fast first draft is worthless if nobody reads it.
The task comes up often, always the same way. AI can handle it, and someone checks now and then.
It’s often the case when
Example: Answering employees’ common questions from the employee handbook.
Watch out: Automating doesn’t mean forgetting. Decide who checks, how often, and what AI must never handle alone.
Nobody should be doing this task. Not you, not AI.
It’s often the case when
Example: A 12-page report sent every month since 2019, with never a reply.
Watch out: Automating a useless task just wastes time faster. Before you drop it, check with whoever asked for it.
They come back for every task. They don’t give the answer. They help the team find it.
A task that comes up often is worth organizing well. A rare task, a little less so.
Time matters, but not on its own. A short, frequent task can cost more than a long yearly one.
This is the risk question. A mistake in an invitation is easy to fix. A mistake in a performance review, much less so.
Some tasks need a person, not just a good piece of writing. If the relationship matters more than the content, the task stays human.
As in the original exercise, every job hides a few traps. Spotting them is part of the game.
A name, a file, a performance review. Before handing a task to AI, ask what you’re allowed to put into it. Often you can remove whatever identifies the person.
A death, a complaint, a conflict. AI can write a lovely message. That’s not the point. The person needs to know that someone is taking care of them.
It looks important, it takes time and nobody uses it. The right bin is not "AI does it".
A rate, a date, a recent rule. AI can produce a number that looks real. It has to come from a source you can cite.
Two documents speak to each other: high overtime and an open position, for example. Connecting them sometimes changes the decision.
After each choice, the game can show the AI’s view. It comes after the discussion, not before, so it doesn’t replace it.
This view is not the right answer. It’s one more opinion. When the team and AI disagree, ask why. Did AI spot a risk you missed? Or is it missing something about your reality?
That habit of judging what AI suggests is exactly what the game is trying to build.
Pick one task to try this week. If an idea looks bigger, the AI Canvas helps you analyse it in nine boxes.