I divide knowledge into three layers.
The outermost layer is tool operation.
For example, where the buttons are, how to write the parameters, how to configure a certain plugin. This kind of knowledge is directly tied to a specific product, interface, and version, so it becomes obsolete the easiest.
The middle layer is methods and workflows.
For example, how to break down tasks, how to cross-check, how to design a human-AI collaboration workflow. This kind of knowledge also changes as technology upgrades, but it does not depend on any single product, so it has stronger transferability.
A deeper layer is principles and judgment.
It includes logic, structure, expression, problem definition, and professional foundations. What they solve is not “how do I press the button,” but “is this worth doing,” “how should the task be broken down,” “is the result reliable,” and “who is responsible if something goes wrong.”
Among these three layers, the largest number of people learn the outermost tool-operation knowledge, especially in the AI era. Because the speed at which new tools appear today is really too fast, and the speed at which tutorials come out is fast too. “There’s no way to learn it all” very effectively covers up the fact that tool-operation knowledge becomes obsolete easily.
Of course, tools should be learned, because they can improve your speed right now, but they do not “hold their value.” If you don’t want to be exhausted from running around nonstop, you should put more effort into the latter two layers.
The middle layer governs how things move forward.
It is not tied to any one product, so it is more durable than buttons; but it still updates with the way work is done, only on a longer cycle.
Take writing a proposal, for example.
Outer-layer knowledge is which tool to open and which template to use. The middle layer is another set of actions: first clarify the reader and the decision purpose, then break the task into research, hypotheses, structure, draft, and review. Change the tool, and this process still holds.
Or take looking at data.
The outer layer is how to click in a certain software. The middle layer is: first define the question to answer, then choose metrics and verify the scope and definition, and only then generate charts. Skip these steps, and the chart may look very nice, but you won’t know what it is answering.
The innermost layer governs whether things are right and whether they are worth doing.
When making judgments, I give you three steps.
First, define the problem.
“Help me increase video views” sounds like a task, but in fact it has no target, no constraints, and no success criteria.
Change it to “for office workers who have followed AI tools but lack systematic understanding, increase retention in the first 30 seconds without increasing shooting frequency,” and then the choice of topic, opening, and tests becomes meaningful.
Then determine the structure, meaning what structure to use to break down the problem.
The same growth problem, if broken down by user journey, by channel cost, or by content structure, will produce completely different task lists.
Finally, use logic and professional expertise for verification.
About logic: the model writes, “After introducing AI grading, scores rose, showing that learning ability improved.” Scores are observations; ability is an interpretation; what’s missing in between is a control group and conditions. If the logic doesn’t work, you can’t see this leap, and then you have no right to verify it.
About professional expertise: people who don’t understand finance can still get AI to produce analysis, but there are too many intricacies in it, and even AI can easily be led astray. The threshold for execution has gone down, but the threshold for verification has not gone down; professional expertise is still required.
Put these two layers together: methods determine whether you can use AI in a stable way, and principles determine whether, after you use it, you will be led around by it.
So, just configure learning by layer: learn buttons when you need them, refine processes again and again, and practice judgment in the areas you are truly responsible for.
What holds its value is not remembering more answers, but that after changing tools, you can still ask the right questions, push tasks forward, and see through the results.