Lately, I’ve been actively exploring the best ways to leverage AI in my day-to-day workflow. I’ve experimented with different models, changed harnesses, created multiple workflows, expanded my domain knowledge into other fields, and even built a second brain.

But I almost overlooked one important thing: mental models.

Mental models are different from domain knowledge. Domain knowledge is what you know about a specific field, while a mental model is how you structure and simplify your thinking to understand situations and solve problems.

Most people try to improve AI output by writing longer prompts. But better results often come from giving AI a clear and simplified representation of how something works in the real world.

The graph shows how the relative usefulness of domain knowledge and mental models changes as a problem moves from familiar to unfamiliar. Domain knowledge is extremely valuable when the problem belongs to a field you already understand. The graph does not mean domain knowledge becomes less valuable over time. It means its applicability decreases as the problem moves outside its original domain.

When you building some product or trying to solve specific problem in your product. There is a common belief that better prompting means adding more words. In practice, a short prompt built around with the right point outperform a long prompt filled with bloated background information. For example, let’s consider a simple prompt:

AI can respond with familiar recommendations: improve onboarding, send reminders, personalize the experience, introduce gamification, or gather user feedback. None of these ideas are necessarily wrong. The problem is that the question does not tell the AI how to examine the situation. As a result, the answer may be broad, predictable, and disconnected from the real cause of the problem.

Now apply this question:

The topic has not changed. The reasoning process has.

The same thing for this prompt:

The request is too generic. It gives the AI a destination, but no framework for exploring the product space. The result will likely be a predictable copy of Trello with standard features such as boards, lists, cards, comments, and notifications.

A stronger prompt would be:

This prompt does more than ask AI to generate features. It gives AI a mental model for exploring alternatives, comparing existing products, challenging assumptions, and creating something more distinctive.

We do not need to become experts in every discipline. However, we should understand a model well enough to know when it is relevant and what a good application looks like. Otherwise, mentioning a mental model becomes another form of prompt decoration. Telling AI to “use first principles” is not useful when the problem mainly requires historical comparison. Asking for “systems thinking” may add unnecessary complexity to a straightforward operational decision. Using inversion for every situation can produce excessive focus on risks while overlooking opportunities.

The value comes from selecting the right mental model for the right problem.

I would not explain all the mental model and tell you which model should we use on every different problem. There are dozen resource out there you could get to learn this topic. But once we understand the purpose of a mental model, AI can help us apply it much better. It can identify assumptions we overlooked, simulate competing perspectives, trace possible consequences, and test whether our conclusion remains consistent under different conditions.

In this relationship, the human chooses the lens. The AI expands the view.

The long-term advantage of AI will not come from memorizing clever prompt templates. For me, templates are useful. i have tons standard template in my project, but they are easy to copy and quickly become outdated. A deeper advantage comes from understanding how different forms of reasoning work.

The future of working with AI is not simply learning how to ask better questions. It is learning how to give those questions a better way of thinking.