Give AI a job, not the whole workflow.

“We should use AI” is an interesting starting point. It is not yet a useful project brief. The next question is more practical: which part of the work needs help, and what would better look like?

I would rather define one useful role for AI than add it to every step. A clear assignment makes it easier to decide what information belongs in the process, what the output should contain, and who needs to review it.

Name the job before choosing the tool

Consider a team preparing campaign briefs. A bounded task might be turning approved planning notes into a first-draft brief. The task is not deciding the audience, inventing product claims, or approving the campaign. Those decisions remain with the responsible people.

Write a short task definition: permitted inputs, expected output, things the system must not decide, and the person who accepts the result. Use only information approved for the tool and workflow. Convenience should not determine what client or customer material gets shared.

Make review part of the work

“Someone will check it” is too vague. Name the reviewer and the checks. For a brief, that could mean checking the audience, supporting evidence, tone, missing context, and unsupported claims. Give the reviewer time and authority to reject or rewrite the draft.

Keep a straightforward fallback. If the output is not useful, the team should be able to finish the task without rebuilding the whole process around it.

Count the work after the first draft

In a small trial, compare the whole task: preparation, generation, review, corrections, and handoff. Ask whether the result is genuinely easier to use, not simply faster to produce. Record where the process still needs human context.

My standard is simple: AI should support a clear purpose without making accountability harder to follow. Give it a defined job, give people meaningful control, and improve the workflow from what you learn.