2026-06-29
Hand reports to AI and growth stops: the line where AI drafts and humans decide
It's tempting to automate reports and evaluation with generative AI, but if AI writes the answers, problem-solving never grows. Keep AI to drafts and hints, and keep the final call human.
The temptation to automate, and the trap
With generative AI you can produce improvement ideas and evaluation comments in an instant. But if AI writes even the answers, members stop thinking for themselves. Convenience quietly takes away the very problem-solving that was supposed to grow — that is the trap of automation. The riskiest case is when the output looks reasonably correct. Because nothing feels off, both the member and the manager tend to submit and approve without scrutiny, and before long no one is reflecting in their own words.
Keep AI to drafts and hints
Limit AI's role to support on stuck days. Instead of filling in the answer, have it return questions like 'where did you get stuck?' and 'how will you change next time?', and the thinking stays with the person. A growth mode where they write it themselves by default, with AI as scaffolding, keeps development alive. As a rule of thumb, have the person first write even a few lines in their own words, and receive questions only if they can't. Seeing an AI draft first tends to pull their thinking toward it.
The final evaluation call is always human
AI suits organizing facts and drafting, but the final call on evaluation and feedback stays with people. Being able to explain who judged on what basis is essential for fairness and accountability. AI's output is material; the decision is human. Even when an AI draft is used for an evaluation comment, the manager should reread it against the facts and put it in their own words before delivering it. It is safest to assume that 'the AI said so' does not hold up as a basis for evaluation.
The line builds trust
Knowing that 'AI won't make the final judgment' brings reassurance to employees and customers alike. AI shortens time, people judge meaning — making that division of roles explicit is what sustains trust in the tool. As an operating rule, summarizing on one page where AI may be used and where only people act helps reduce hesitation.
A minimal first step to try
Rather than rolling this out to everyone at once, start with your own reports or a few members, in the form 'the person writes first and receives questions only when stuck.' After a while, review whether their writing has grown shallow or the manager's judgment has drifted toward the drafts, and if so, narrow the line. Adjusting this way is the realistic approach.
Related articles
The feature that fits this article's topic: AI assist →
A tool for a culture of improvement and fair evaluation that implements these ideas.