When AI helps us run tasks automatically, work often needs to be handed over. The first half is done by one AI, and the second half is handed over to another, sometimes to a different company, a different machine, or even a different person. Every time we hand over, we need to tell the person taking over what to do next. This sentence is the handover instruction. This article will tell you that the main reason for a bad handover instruction is talking too much, and it will give you three ways to write it for different scenarios and a handover form you can directly copy.
People who let AI automatically run tasks and often hand over to another AI or sub-agent; creators who use multiple AIs for different tasks; advisors or independent workers who use the same AI tools; people who lead a team or family to use the same AI tools.
A core principle: Only write what the other person needs, not what I will do later. Three ways to write handover instructions for different scenarios: one-time assignment, supplier model, and handover to another person to continue. A handover form you can directly copy, and the right amount of feedback when things get stuck.
Real-life example: I wrote three extra sentences, and I got called out
Once I was making a webpage and needed two images. Since I couldn't come up with them on my own, I handed them over to another AI to generate.
I wrote the handover form quite carefully. Each image needed to have what content, what ratio, where to save it, and how to report back to me after it was generated. All of these were correct.
Then I added three more sentences on the fly: the images should be placed after which paragraph in the article, the desktop version shouldn't be too wide, and the mobile version shouldn't crop the text.
I got called out. Those three sentences shouldn't have been in the image generation handover form. The person generating the images only needed to generate the image files. How to place the images in the webpage and how to adjust the layout were tasks I needed to do myself after receiving the images. I forced myself to do what I needed to do later into the handover form for someone else.
Why do we write too much?
After I finished writing, I thought back about why I wrote those three sentences. There were three reasons, and I think many people are like me.
Want to explain more, explaining while explaining, ended up explaining things that were not his responsibility.
Unable to tell which tasks belong to the other party and which are my own. Everything in my mind is connected, and I poured it all out at once.
Sent the complete imagination of the finished product all at once. However, the other party only needed a small part of it, and the rest was noise to them.
One principle, three scenarios.
Core principle: only write what the other party needs.
Take the example of a drawing. A handover form only needs four things: what the image should have, what ratio, where to save it, and how to report after generation. Layout, position, and where to put it in the file are all deleted, as those are my own matters.
Three types of handover, each with its own way of writing.
Hand over the whole task, and let the other party complete it. Write clearly which files can be modified, which cannot, what state counts as completion, and which command to use for verification. The core is the whitelist; anything not written can be modified is strictly prohibited from being touched.
The main line needs a part I cannot generate myself, which I outsource. After the other party generates it, they report where the file is and the specifications, and I put the part back into the main line. The handover form only writes the part specifications, not where to put it or how to assemble it.
Switching to another person with the same account and data, like moving work from a desktop to a laptop or handing it over to a family member. The handover instructions should first clarify who is currently in the conversation. Reporting should be in plain language, and things like pricing strategies should be pre-recorded for me to approve later.
The balance of automation: only report when stuck
I want the process to run automatically. Mechanical tasks should be handled by cheaper, faster hands. For image generation, directly assign it to another party. These don't need to bother me. Only when it truly gets stuck should we stop and provide me with a handover form, which I will manually transfer to another party. Once that party completes the task, they should provide a list for me to paste back.
Preventive measures: three common ways cheap hands fail
- Imagine it's done. Reported it as completed, but actually did nothing. The handover form includes a command that can be directly executed for verification. I run it again myself and don't take their report as the truth.
- Expand the task on the fly. Do the things that look like they should also be done together. First write what this round does not include, so they don't treat good ideas as part of this round's task.
- Overstepping authority. Do my downstream tasks and include them. Only write what the other party needs, and delete any extra sentence.
What it guarantees and what it does not guarantee
How to start
First copy this handover form, and apply it to image generation, verification, and outsourcing a segment of material.
Then do one thing: pick a task you recently handed over incorrectly, spread out the handover form, and use the deletion method to cross out each line of 'things I will do myself when I return'. You will find that the remaining lines are what the other party really needs.
Good handover instructions often save more than what is written. The form you write for others, the thinner it is, the more accurate it usually is.
Use AI as a tool to help you accumulate judgment
I host two free online seminars every month, sharing how to organize workflows, judgments, and experience into prompts, skill packages, and knowledge bases that AI can flexibly use. If you want to continue learning or have consulting needs, feel free to start from the community.
AI × Knowledge Management, Tacit Knowledge Distillation, Together to Use AI as a Thinking Partner.
Join the community ↗