Design a scenario with protection: person at the control point + clear boundaries. Return JSON {"scenario_name":"...","trigger":"...","ai_step":"...","human_checkpoint":"...","do_not_automate":"..."}. ai_step — what the AI is preparing; human_checkpoint - where and why the person approves the decision; do_not_automate - what do you deliberately leave for people in this process and why (the cost of error).
Maturity in automation is knowing where to stop. Don't automate things where the cost of error is high and irreversible: money, legal decisions, layoffs, public statements, medical and personal topics. Don't automate the rare and unique - customization won't pay off. Don’t automate what you can’t describe with a rule: if you decide every time “by feeling,” the machine will make the same mistakes, only faster and silently. Safe design mechanics: place a person at a control point where the decision is important or irreversible - AI prepares, the person approves with one click. This is a hybrid: machine speed plus human common sense. Leverage: Ask yourself 'what's the worst that will happen if this move goes wrong?'. Cheap and reversible - automate with confidence; expensive and forever - leave the checkpoint. Insider: automate the preparation, not the final click - 90% of the routine is collection and drafting, and let the decisive click remain with the person. Second trick: create 'safe words' and limits - if the AI sees a mention of money, a complaint to the court or termination, it does not act on its own, but calls the person. Typical mistake: automate for the sake of beauty. If it’s faster and more reliable to use your hands, use your hands.
Unlock access to submit solutions for instant AI review.