AI Task Map

List the work you actually do, sort each task with three questions about judging, context and stakes, and settle the unclear cases with a small test on material where you already know the answer.

Mixed research: three studies, each in one setting, find large gains on some tasks and worse results on others (Noy & Zhang, 2023; Brynjolfsson, Li & Raymond, 2025; Dell'Acqua et al., 2026). The three sorting questions and the delegate, collaborate and keep-away split are practice, not research.

Use it when:

• You use AI tools here and there and want a clear view of where they belong in your own work.
• Your team is about to write an AI policy or an adoption plan and needs a task-level picture first.
• A new model or tool has arrived and you want to know which of your earlier verdicts still hold.

Skip it when:

• You have not yet used an AI tool or do not know the basic terms: start with 10.1 AI Concepts Map.
• You already know the task and need to write a good request for it: use 10.8 Prompt Brief.

Whether AI helps depends on the task, not on the job. Three careful studies show the same shape in different settings. Professionals writing short, self-contained texts finished faster and wrote better with an assistant, and weaker writers gained most (Noy & Zhang, 2023). In a large support center, an assistant raised the pace of the least experienced agents most, while the most experienced barely sped up and lost a little quality (Brynjolfsson, Li & Raymond, 2025). In a consulting experiment, help on tasks within the tool's reach raised speed and quality, but on one task chosen to lie just beyond it, people using the tool reached the right answer less often (Dell'Acqua et al., 2026). The authors call that uneven edge a jagged frontier: it does not follow how hard a task looks.

The limit is that each study covers one setting and models that have since changed. So this page does not hand you a list of good AI tasks. It gives you three questions to sort your own, and a test for the cases where you cannot tell.

Write tasks at the level of a piece of work you could hand to someone, not a role or a whole process. Add the hours a month, roughly, so you see where a gain would matter.

Strong answer: "Write the monthly comment on cost variances for the plant manager, about 6 hours a month (illustrative)." Not: "Finance work."

Copy rows for more tasks. Five to eight is enough for a first map (rule of thumb).

Illustrative example: the role, the company and all numbers are made up.

A financial controller at a sawmill group in Finland maps her work before the team writes its AI rules.

Section 1: seven tasks, from meeting notes (about 4 hours a month) to the monthly variance comment (about 6) and the board memo draft (about 8).

Section 2: meeting notes are quick to judge and low stakes: delegate. The board memo is quick to judge but carries her name: collaborate. The variance comment needs figures that must not leave the company and a subtle error would mislead the plant manager: test first. Final judgment on inventory write-downs: keep away, it is her call and her accountability.

Section 3: she rebuilds an old variance pattern with invented figures. The tool writes a convincing comment but reads a timing difference as an overrun. Verdict: collaborate, with every figure recomputed. A second test, summarizing a public industry report, passes.

Section 4: two tasks delegate, four collaborate, one keep away. She sets a retest for when the company's approved tool is upgraded.