People say telling AI what to do is difficult. But asking another person involves explaining, schedules, emotions, trust, and responsibility. Sometimes the human is the harder one to work with.
This connects two things: people who use AI only as a search box and conscientious workers who end up doing everything themselves. Both have trouble seeing how work can be handed over.
1. The boss goes to bed
Imagine the boss saying, “I'm going to sleep. Handle everything else.” Say that to human employees and an uncomfortable meeting may follow. AI can help with research, ideas, writing, translation, simple programs, and revisions.
But closing a chat does not keep it working forever. Continuing tasks requires schedules, connected tools, permission, and checks. The boss's wonderful sleep comes from preparation, not magic.
2. A few thousand yen for many roles
One person may spend roughly ¥3,000 a month, while heavier use could involve a budget of ¥16,000–¥30,000. These are examples of spending, not a verified current price list for any product.
Trying work that previously needed several specialists can feel extraordinarily cheap. Still, subscriptions do not include all the costs of review, corrections, or mistakes. “The AI seemed confident” is not a quality certificate.
3. A powerful tool becomes a paid search box
Asking AI for information is useful. But if that is all you imagine, you miss writing improvements, comparisons, organizing data, brainstorming, and simple prototypes.
It's like hiring a giant moving truck to deliver a single eraser. Job completed. Eraser safe. Truck now questioning the meaning of its existence.
4. Not knowing what to ask is a valid starting point
People who rarely use AI are not necessarily unimaginative. They may lack time, permission, trust in its answers, or an example relevant to their lives.
“I don't even know what my problem is. Ask me questions so we can figure it out” is a perfectly good request. AI can help form the question before answering it. You do not need a polished instruction.
5. Humans can be harder to delegate to
A colleague has their own schedule, workload, feelings, opinions, and relationship with you. Ten rounds of “No, start over” can damage that relationship.
AI is easier to ask for repeated drafts. Humans, meanwhile, understand hidden context, care for others, build relationships, and accept real responsibility. This is not a contest with a universal winner. The coordination burdens differ.
6. Why responsible people work overtime
A careful worker delegates. Explaining takes time, then checking takes more. If something breaks, the worker fixes it and apologizes. Eventually: “It'll be faster if I do it myself.”
So they do everything, stay late, and hear, “You should delegate more.” Fantastic idea. Could somebody delegate the work of delegating?
This is not just personality. Limited authority, busy teammates, no time for training, and responsibility without control all make delegation costly.
7. Subscription sold, meaningful use not achieved
Imagine persuading someone to pay for AI. Then you explain its features and hear, “Hmm. Sounds hard.” Sale completed; actual use still pending. Congratulations on installing a luxury search box.
A hundred features can sound like a hundred new chores. Solving one annoying task for that person may explain the benefit better than an impressive list. Paying for access and changing a habit are separate achievements.
8. Research shows both benefits and limits
In a 2025 survey of US employees, 10% reported using AI at work daily. This is about that US sample at that time, not the whole world.[1]
In a product-ideas experiment, one person with AI performed about as well as two people without it—but only for those kinds of tasks.[2] Another experiment involving about 6,000 workers found that actual AI users spent about three fewer hours each week on email, without a clear drop in meeting time.[3]
A customer-support study reported a 15% average productivity improvement.[4] Conversely, experienced developers in an early-2025 experiment took 19% longer with AI. Follow-up observations in 2026 had selection problems and could not firmly measure the newer effect.[5] The task and review costs matter.
9. Give it one problem instead of a perfect question
Describe a recurring frustration. Ask AI to divide the task into steps and ask what information is missing. Let it handle one low-risk draft or sorting job, and check the output before expanding.
“Our handovers are always chaotic. What should we record?” is enough. For humans and machines alike, set the deadline, definition of done, and reviewer. Never enter confidential business information into unapproved tools or let AI decide serious contracts or personnel matters without proper human authority.
10. A good company lets everyone rest
Searching with AI is not wrong. But there is another loop: imagine, ask, inspect, revise. Knowing how to hand work over, and how to verify it, opens far more possibilities.
People who say “I'll do it myself” need time and authority to delegate, not another lecture about efficiency. Otherwise checking employees' work is simply replaced with checking AI's work.
The ideal company is not one where only the boss sleeps. Everyone should be able to rest. Boss, before bed, please at least write down what “done” means.
References (5)
- Gallup, AI Use at Work Rises (2025) gallup.com
- Dell’Acqua et al., The Cybernetic Teammate (2025) nber.org
- Dillon et al., Shifting Work Patterns with Generative AI (2025) microsoft.com
- Brynjolfsson, Li & Raymond, Generative AI at Work (2025) academic.oup.com
- METR, Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity (2025) and 2026 follow-up: ; https://metr.org/blog/2026-02-24-uplift-update/ arxiv.org
