What Tasks Should You Hand to AI? A Four-Question Test

You have killed the work that should not exist. What survives is a pile of necessary tasks, some of which need you and some of which do not. The skill now is sorting them correctly, because handing the wrong things to AI wastes effort, and keeping the wrong things wastes the tool. Four questions do the sorting.¹

Question one: can you verify it?

This is the most important filter, full stop. If you have no way to tell whether the output is right, do not hand it over, because you will simply be laundering uncertainty into confident-looking text. AI is a machine that produces plausible answers, not necessarily correct ones, and the only thing converting plausible into trustworthy is your ability to check. No verification, no delegation.

Question two: is it worth the brief?

If explaining the task properly would take longer than just doing it, do it yourself. A one-off task with a complex, hard-to-articulate goal is often faster done by hand than described to a machine. The exception is anything you do often: there, a reusable brief pays for itself many times over, so the upfront cost of explaining it well is an investment, not a waste.

Question three: is it outside your core craft?

Hand over the supporting and the shallow, not the differentiated work you are actually paid for. Your core craft is both your value and the thing that keeps your skills sharp, so delegating it is a double loss: you weaken your edge and you stop practising. Keep the judgment, taste, and original thinking. Give away the scaffolding around it.

Question four: how will you check it, and how hard?

Decide your verification plan before you delegate, not after, and match its depth to the stakes. A throwaway internal summary needs a glance. A figure going into a board paper needs a full check against the source. The question is never whether to verify, only how much.

What are the good and bad candidates?

With those filters, the candidates almost name themselves. Safe to delegate, in rough order: first drafts you will edit, summaries, reformatting and translation between formats, pulling structure out of messy material, routine research you will fact-check, brainstorming where you want volume and will pick the winners, and boilerplate of every kind. Keep for yourself: anything you cannot verify, anything high-stakes, anything outside your knowledge where you could not catch an error, and your core craft.

What is the move most people miss?

The most powerful delegation is often not asking the machine to do the task, but to build the thing that does the task. Faced with a thousand records to sort, the junior move is to ask AI to read them one by one. The senior move is to ask it to write the script that sorts them.² Use AI as a toolsmith, not a grunt. Delegate the meta-task, and one good brief can do a thousand jobs while you do something better with the afternoon. The point of all this sorting is not to offload as much as possible. It is to offload the right things, so the time you reclaim is real and the work that carries your name is still genuinely yours.

Frequently asked questions

What tasks should you delegate to AI?

Hand over work you can verify, that is not worth the time to brief in detail only if rare, that sits outside your core craft, and that you have a plan to check. Good candidates include first drafts, summaries, reformatting, routine research, and brainstorming.

What should you not delegate to AI?

Keep anything you cannot verify, anything high-stakes, anything outside your knowledge where you could not catch an error, and your core craft, the differentiated work you are paid for that keeps your skills sharp.

What is the senior move in delegating to AI?

Often the best delegation is not asking AI to do a task but to build the thing that does it. Faced with a thousand records, ask AI to write the script that sorts them rather than reading them one by one. Use it as a toolsmith, not a grunt.

About the author

Tom Goodwin is the author of Don’t Work Harder, a book about taking the time AI gives back as time rather than more work. He is a co-founder of GAMEPLAN and writes on productivity, technology, and the economics of the working week.

Footnotes

Adapted from practitioner decision frameworks for delegating to AI: verifiability, the effort ratio of briefing versus doing, whether the task is core, and a stakes-based verification plan. ↩
The “toolsmith, not grunt” reframe: having AI build the tool that does the task rather than grinding the task itself, so one brief scales to many jobs. ↩