Stop doing everything yourself: hand over the work to the AI ​​Legion’s six-step workflow

After working for a long time, you will find that the real bottleneck is not the inability to use AI. You have touched the tools and written prompt words, but at the end of the day, things still pile up on you alone.

The problem is that most people regard AI as a very smart pen, and they can use it temporarily when they need it. And one thing I have become more and more certain of as I practice this whole year is: Use AI as a team, not as a tool. For a team to be able to operate, it must have a structure, a division of labor, and a system. It cannot start from scratch every time.

In this article, I break down the task of handing over work to the AI ​​army into six steps that can be followed. And for every step, I made a corresponding small tool in my laboratory lab.vista.tw (opens in a new tab), most of which are free and can be used immediately after opening. You don’t have to just read the concepts, you can build your own workflow while reading.

These six steps are a line like this:

1. Before starting work, check your blind spots.
2. Pour out all the work and organize it into an army
3. Quickly separate what should be paid and what should be kept.
4. Make an in-depth judgment on the most difficult thing to decide.
5. Write down the rules scattered in your mind into a system file
6. Go back and cut off those subscriptions you can no longer afford.

Step one: Before starting work, check your blind spots.

Many people’s frustration with AI is that “it still doesn’t work after correcting it ten times”, and then blame the AI for being too stupid. But the more common truth is: it’s not that it’s stupid, it’s that you missed something. You know in your mind who your audience is, which bottom line you must not touch, and what the format should look like, but you don’t say it out loud, so the AI ​​can only guess.

Therefore, the first step in the entire workflow is not to rush to assign tasks, but to take stock of your own blind spots. This also echoes what I have always advocated: The starting point for making full use of AI is not to pick tools, but to dismantle the workflow. Spreading out the tasks and checking the things “I know but haven’t said it yet” one by one is much less labor-intensive than correcting them ten times afterwards.

I made this thing into a free tool 〈Blind Spot Patroller (opens in a new tab)〉: spread the task into four quadrants, check the audience, format, bottom line, and limitations that others often forget to mention, watch the blind spot coverage climb from the complete blind area to “inspection ready”, and finally take away an AI prompt that will ask you key questions before starting work.

Step 2: Pour out all the work and form an army

After taking stock of yourself, next take stock of your work. A one-man company does not do everything by itself, but knows clearly what should be retained and what should be handed over. But “Leave it to AI” is not a slogan. You must first know which roles you need.

I made 〈AI Legion Organizer (opens in a new tab)〉 to help you take this step: pour out the daily tasks, and judge one by one whether you should keep it yourself, leave it to AI, or human-machine collaboration, and then match eight AI character prototypes to produce your one-man company legion organization chart, and even give you the system prompts for each character’s starting moves. This is the entry-level version of the same logic as the army I described in “Three months, one person, an army named Claude Code”.

Step 3: Quickly separate what should be paid and what should be kept

With an organization chart, new tasks will continue to pop up every day. Every time something comes up, I have to struggle again with “Should I do this or lose it to the AI”, and I will soon get tired. You need a sorting action that is so fast that you don’t have to think about it.

〈AI Delegation Map (opens in a new tab)〉 is used for this purpose: put the work on hand into the four quadrants, and get a list of “should be handed in vs. should be kept” in five minutes. It does not pursue thoughtful thinking, but rather allows you to develop a reflex: when a new task comes in, first ask, “Which category does this item belong to?”

Step 4: Make an in-depth judgment on the most difficult thing to decide.

After the list is divided, there will always be a few items stuck in the middle that make you feel uneasy no matter how much you think about it. Carrying everything on your own is the biggest ceiling for growth; but if you hand over some things, you are afraid of something going wrong. At this time, it is better to rely on a set of signals than to rely on feelings.

〈Should this matter be handed over (opens in a new tab)〉For a specific job, eight signals are used to score each item to determine whether it should be kept by ourselves or handed over to others, outsourced or AI, and the risks that need to be compensated before handing over are marked, and finally a “handling instruction manual” is produced that can be sent out directly. This is what struck me most when I read “Inverse Division of Labor”: When a person starts to own an AI army, what you have to practice is not to be better at doing things, but to be better at judging what should be handed in and how to hand it over.

Step 5: Write the rules scattered in your mind into system files

After completing the first four steps, you are already assigned. But if you have to re-explain “our rules” verbally every time, then you are still the bottleneck. Real amplification occurs when you take the rules out of your head and write them into a system that the AI ​​will read before every start of work.

I was particularly impressed by this when I was working on an AI editing system: If you only rely on conscious awareness of process disciplines, you will fail sooner or later, and you have to rely on writing process codes and system files to maintain them. 〈AI Legion Commander Console (opens in a new tab)〉 is to help you implement this step: set your command philosophy, routing index, delegation template and model schedule, and download four system files that can be directly fed to the AI ​​to read with one click, so that each AI knows the rules before starting work. This is a key step in upgrading from “being able to use AI” to “commanding an AI army”.

Step 6: Go back and cut off those subscriptions that you can no longer afford.

The last step is often overlooked, but it is very healing. When the AI ​​army begins to take over the work, you will find that a bunch of tools that were previously ordered to support the process are actually no longer open. The hand that pays is always faster than the hand that cancels.

〈AI Subscription Discontinuation Physical Examination (opens in a new tab)〉Help you pour out your subscriptions, answer honestly “When was the last time you really used it?”, automatically determine the three lists of retention, observation and elimination, calculate how much you can save in a year, and also take away an AI financial coach prompt to question you one by one. Leave your budget to the tools that actually help you fight the war, and this army will last longer.

These six steps are actually an operating system

After walking through this circle, you will find that these six tools are not six independent gadgets, but a complete line: taking stock of blind spots, forming a legion, daily triage, in-depth judgment, writing down the system, and replacing tools. Together, they are an operating system that enlarges one person into one person plus an army.

What you save seems to be time, but actually it is willpower. By handing over repetitive losses to the process, you can retain the energy to decide where to go.

In this workflow, several tools (assignment maps, assignment instructions, and commander consoles) are currently available to subscribers and will not be made public until a set date. If you want to use it as soon as possible, and also want to regularly receive my first-hand experience of experimenting with AI in a one-man company, you are welcome to subscribe for free:

---

This entire set of tools lives at lab.vista.tw (opens in a new tab), a laboratory built by myself using AI. It is divided into five groups according to identity. You can choose a group for your own use. I usually put more ideas for creators and independent workers at content.tw (opens in a new tab), and you are welcome to visit vista.tw often.