I think ordinary people need to think more systematically about how to approach AI.
Take a company as an example. In the past, we organized work around positions. Now that we are in the AI era, I think an AI-native company can distinguish and arrange work more around tasks.
In fact, we can first define something called a role.
For example, I am both a father and a son: I have parents above me and children below me. And I am also the owner of a company.
Under each role, there are actually many tasks.
For example, as the owner of a company, I need to assign and communicate tasks to employees. That also requires the employees themselves to have some understanding of AI.
We currently have around six employees. By my own sense, AI’s share of the work has basically reached 80 percent. In other words, for the vast majority of things—whether R&D or other kinds of work—we have handed them to AI.
This even includes the quality-inspection app on our production line, as well as checks within the quality-inspection process. We use AI for those too.
For example, we can ask it to look at the entire quality-inspection line: which processes are wrong, and which parts need adjusting.
So we are a company that uses AI very heavily.
But that is not the main point I want to discuss today.
What I want to say is that I think everyone can start by listing out their own tasks.
First, plan out what your tasks are for this week, this month, and the future.
And by tasks, I do not mean only work-related tasks. Some may go beyond your current job role.
For example, this is the era of the “super individual.”
If you work in sales, for instance, you can also run your own Xiaohongshu, WeChat Channels, or Douyin account, and talk about the professional knowledge you know from your role.
Because we now have AI, we can continue to develop these things more deeply.
So today I will share some of the tables I use and put them in the video for everyone to look at.
For example, in my company, we now manage work entirely by tasks: work is arranged one task at a time.
For these tasks, I make a clear distinction: which ones AI can execute, which ones AI cannot execute, and which ones require people to monitor and check during execution.
For example, for certain hardware inspections and QC, someone must be on site.
AI can help you review the process and find problems, but some specific actions still have to be done by people.
So I have a grading framework for AI execution.
It asks: if AI is doing this task, where are the people?
Some tasks can be closed out by AI.
For some, AI completes the first pass and people review it.
For others, people do the work and AI assists.
For example, in an interview, you communicate with a candidate face to face. People still do that process, while AI provides support within it.
Then, for every task, I list the task line, task name, specific actions, deliverables, acceptance criteria, and execution frequency.
In this way, you know exactly what you want AI to help you accomplish.
Otherwise, you may only give it a task, while not having thought clearly about what level of completion counts as doing it well.
There is another issue: AI agents actually go through a process of being trained.
During execution, they can sometimes make you impatient, because they may consume a lot of time. A large part of your time may be spent on debugging.
During that debugging process, you also need to gradually adapt to this way of working, and clarify the tasks and requirements bit by bit.
So I think you can start with your own roles, list out your tasks, and then make clear what AI does and what people do within each task.
That is how I am working now.
