Hi everyone, I’m KunYuan. I’m building an AI company and conducting research in Singapore, and I’m gradually turning this Substack into RainbowCity | Rebuilding Your Cognitive OS—a philosophy community for the age of algorithms and AI.
In Essay 18, I wrote about why philosophy is the antidote to existential anxiety. At the end of that essay, I arrived at a conclusion that matters deeply to me: if philosophy remains only at the level of understanding, it is still not enough. Algorithms and AI are training us every day. Recommendation systems train our attention. Metrics train our expression. AI tools are making us increasingly accustomed to shortening—or even skipping—the process by which judgments are formed.
If external systems are continually shaping our cognition, then philosophy must become part of our everyday lives—a capacity we deliberately cultivate through repeated practice.
This is the question I want to explore today: What is Cognitive OS?
I know the term may sound like a new methodology, a productivity system, an AI-era workflow, or even a more advanced form of knowledge management. But I want to distinguish it from all of these at the outset.
Cognitive OS is not designed to help us produce more, process more information, or simply write better prompts. It addresses a more fundamental question:
In an age when AI can rapidly summarize information, generate content, rank options, optimize outputs, and make recommendations, do we still retain the capacity to understand the world, form judgments, initiate action, and revise ourselves in light of reality?
In other words, Cognitive OS is concerned not only with whether we can arrive at answers faster, but with the extent to which those answers have genuinely passed through our own understanding, scrutiny, and judgment. The question is not whether AI has participated, but whether human cognition remains genuinely present in the process.
Each of us already operates with some form of default Cognitive OS. Family, education, language, culture, relationships, institutions, platforms, algorithms, and the consequences of our past actions all shape what we notice, how we understand, how we judge, and how we choose. Most of the time, however, this architecture operates implicitly. We rarely see it clearly or reflect on it deliberately.
So when RainbowCity speaks of Rebuilding Your Cognitive OS, it does not mean imposing a single standardized way of thinking on everyone. It means bringing these normally hidden cognitive structures into awareness, so that we can see them, examine them, reconstruct them, and continually update them in response to reality.
This is where Cognitive OS begins: bringing philosophy back into the real processes of cognition and action that unfold every day.
We live in an age that tends to recast every problem as a problem of efficiency.
Too much information means we need better knowledge management. Too many tasks mean we need better time management. More powerful AI means we need better prompts. Greater pressure to produce means we need smoother workflows. A more complicated life seems to demand a more advanced system capable of helping us handle even more.
All these tools may be valuable, and I am not opposed to them. A good note-taking system, a clear workflow, and the right AI tools can genuinely reduce confusion, free up time, and make difficult work easier. But no matter how powerful these tools become, we must preserve one non-negotiable principle:
Tools may help us carry out tasks, but they must not remove human judgment from the process.
Productivity systems usually ask: How can I complete this task faster? How can I process more information? How can I reduce friction? How can I make my output more consistent? These are all important questions. But they often assume that the task is already worth doing, that the problem has been framed correctly, that the direction has already been chosen, and that the relevant standards are already sound.
Cognitive OS asks questions further upstream: Is this task worth doing at all? Who defined this problem? Why should this information be trusted? By what standard is something being called “optimal”? Did I genuinely participate in forming this judgment? What ordering of values underlies this decision? Does this expression still belong to me? Am I willing to bear the consequences it creates?
This is the fundamental difference between the two. Productivity systems primarily optimize external execution; Cognitive OS organizes and protects internal cognition. Productivity systems seek to reduce friction. Cognitive OS reintroduces the pauses, comparisons, and deliberation that genuine judgment requires.
For this reason, I prefer to think of Cognitive OS as a form of infrastructure for cognitive sovereignty. First, it serves as a line of defense, preventing us from unconsciously surrendering the authority to frame problems, rank values, and form judgments as we move through technology’s frictionless pathways. But it is more than a defensive mechanism. It also helps us generate new understanding, form positions of our own, translate judgment into action, and revise our understanding in light of the consequences of what we do.
What it protects is not the principle that every part of the work must be done personally. It protects something more fundamental: Do I still know which tasks can be delegated, which judgments I must form for myself, and which responsibilities cannot be outsourced?
When AI helps me write a paragraph, is it polishing a judgment I have already formed, or generating a position on my behalf? When AI summarizes a document, is it organizing information, or has it already decided what matters and what can be ignored? When AI ranks several options, is it broadening my perspective, or quietly setting the value criteria for me?
If we stop asking these questions, it becomes all too easy to surrender our judgment in exchange for a smoother process. The first function of Cognitive OS is to make these questions visible again.
One of AI’s greatest powers is that it continually shortens the path from question to answer, and from intention to action.
We ask a question, and it answers immediately. It generates text, and we publish it. It ranks options, and we select one. It summarizes material, and we accept the conclusion. It makes a recommendation, and we click “confirm.”
This is undeniably convenient, and it saves enormous amounts of time. But as Essay 17 argued, AI’s deepest risk is not merely that it may produce incorrect answers. It is that it allows us to experience less and less of how those answers are formed. One of the central mechanisms of Cognitive OS is therefore to recreate a cognitive interval within these increasingly short and frictionless pathways.
This interval is not meant to slow us down for its own sake, nor is it a reason to reject AI. It is a necessary pause that allows human cognition to re-enter the process. Before using AI, I can ask: Why am I asking this question? Has the question been framed properly? While interacting with AI, I can ask: How is it changing the boundaries of the problem? Which variables are being included, and which possibilities are being pushed out of view? After receiving an output, I can ask: Is the evidence reliable? Do the reasons hold? Who set the criteria—and according to whose values? Which uncertainties have been concealed? After taking action, I must continue asking: Do the real-world results support the original judgment? Do I need to revise the answer—or the way I arrive at answers?
The cognitive interval, then, does not exist only between an AI output and a person’s final action. It can appear before the question is asked, during the interaction, after the output is produced, and after the consequences of our actions return to us.
The cognitive interval is not Cognitive OS itself. It is the operating space that allows Cognitive OS to re-enter the process. Within that space, clicking “confirm” is no longer merely a formal act. It becomes a process involving understanding, deliberation, choice, and responsibility. When this interval remains open, AI can genuinely augment our cognition. When it disappears, we can easily be relegated to the end of the workflow, reduced to an interface responsible only for authorizing, confirming, and bearing the consequences.
A complete Cognitive OS extends far beyond these four practices. For everyday training, however, Evaluation, Judgment, Decision, and Expression offer four of the most direct and accessible points of entry.
They are not four isolated modules, nor do they form a pipeline that can operate in only one direction. New evidence can change a judgment. Expression can clarify a position. The consequences of action can force us to evaluate the problem again.
I begin with these four practices because they arise almost every day as we use AI, read information, make choices, or speak in public.
The first entry point is Evaluation. Evaluation does not mean gathering more information. It begins by guarding the gateway through which information enters our understanding.
We encounter vast amounts of content every day, and we are becoming increasingly accustomed to letting AI read and summarize it for us. AI can rapidly extract key points, compress material, organize arguments, and turn complex texts into a few clear paragraphs. This is extremely useful.
But if we read only AI-generated summaries, never inspect the sources, never trace the evidence, and never know what has been omitted, then our judgment is built from the outset on a world that has already been organized for us.
Evaluation asks:
How was this question framed? Where did the material come from? Is the evidence reliable? Is this answer stating a fact, or offering an interpretation? Has a particular position been presented as a neutral summary? What has been amplified, and what has been omitted?
This does not mean that we must doubt everything or trace every claim back to its source on every occasion. No one can personally verify all available information. We inevitably rely on summaries, experts, tools, and the work of others. But we must preserve at least one basic awareness:
An AI summary is not the world itself. It is a particular selection, compression, and organization of the world.
Suppose we ask AI to summarize an article and it gives us three clear conclusions. It is tempting to adopt them immediately. But the summary may have omitted the conditions and caveats attached to the author’s claims, weakened the counterevidence, or compressed a complex argument into an answer that is too smooth. What we eventually respond to may no longer be the original problem, but a version reorganized by the system.
The practice of Evaluation does not require us to reject summaries. It requires us to retain sovereignty over the information gateway before and after summarization. We can ask AI to extract key points, but we should know what they are based on. We can ask AI to compare materials, but we should trace the sources. We can accept compression, but we must remember that compression always entails selection and omission.
The ultimate question of Evaluation is: What, exactly, am I basing my understanding of this problem on?
Evaluation is a crucial entry point within Cognitive OS. Without it, even the clearest judgment may rest on a cognitive terrain that the system has already arranged in advance.
The second entry point is Judgment.
Judgment is not simply choosing the answer that appears most reasonable from a list of options. It means forming a position at the intersection of evidence, reasons, models of reality, and values—a position I understand and am willing to take responsibility for.
AI is exceptionally good at generating reasons. Give it a question, and it can produce arguments for and against, write a seemingly balanced analysis, and construct a complete, clear, and persuasive chain of reasoning. These reasons can broaden our perspective and reveal problems we had not previously seen. But frictionless reasoning is not necessarily sound reasoning.
What assumptions does a reason depend on? Are there counterexamples? What costs does it ignore? Does it conceal a particular ordering of values beneath the surface of “rational analysis”? Is it merely the explanation that is most common in the data, superficially plausible, and easiest to accept? Judgment begins at this point.
If Evaluation focuses on the question, “What am I relying on?” then Judgment goes further and asks: Why am I willing to form this position on the basis of this evidence and these reasons?
Judgment requires us to search for counterexamples, examine assumptions, compare different interpretations, and bring our own value anchors back into the process.
AI may tell me, for example, that one career path is better because it offers higher income, faster advancement, and stronger market demand. All of these reasons may be valid. But I must still ask: Is this a direction to which I genuinely want to devote my finite time? What does it require me to sacrifice? Does it align with how I order life, relationships, and meaning? If the choice brings long-term pressure and costs, am I willing to bear them?
AI can help me see more reasons, but those reasons alone cannot form a judgment that is truly mine.
The practice of Judgment turns system-generated reasons into a position I have examined and am willing to take responsibility for. In the end, I must know why I stand where I do—and what new evidence would lead me to revise that position.
The third entry point is Decision.
Decision and judgment are closely related, but they are not the same. Judgment asks: What understanding of the situation do I believe is justified? Decision asks: In the presence of uncertainty and cost, what action am I prepared to take?
AI is highly effective at laying out options. It can analyze the advantages of Plan A, the risks of Plan B, and the costs of Plan C. It can build tables, simulate outcomes, rank priorities, and recommend what appears to be the optimal path. But real-life decisions are rarely just about finding an objectively optimal solution.
We rarely possess all the information, and we cannot eliminate every risk. Choosing one direction means giving up another. Committing time to one pursuit means being unable to commit it elsewhere. Choosing speed may require sacrificing depth. Choosing safety may require sacrificing creativity. Choosing growth may come at the expense of genuine relationships.
The core of Decision, therefore, is not finding the smoothest path. It is committing to a direction that carries real costs under conditions of uncertainty.
AI can help us identify options, but the standards of comparison must remain visible. It can simulate outcomes, but the hierarchy of values cannot remain hidden inside the system. It can alert us to risks, but it cannot bear those risks for us.
When making a decision, Cognitive OS asks us to pause: By what standard has this option been called “optimal”? What does it sacrifice? Which costs can I bear, and which boundaries must not be crossed? If the result is disappointing, am I still willing to acknowledge that this was a choice I made after understanding it? What kind of person will this decision gradually make me?
These questions may slow the choice down, but they also restore weight to it. Often, what we most want to escape is not the range of options, but the cost and uncertainty behind the act of choosing. AI can make the options clearer, but it cannot eliminate the burden that remains ours.
If we simply accept the path AI recommends and carry it out, decision-making gradually deteriorates into mere compliance with system-generated advice. But a human life cannot be reduced to carrying out a sequence of system-generated recommendations.
Through choosing, acting, and bearing the consequences, we gradually shape who we become.
The fourth entry point is Expression.
Expression is not text generation. This distinction matters enormously in the age of AI, because AI is extraordinarily good at producing language. It can make a passage clearer, more polite, and better structured. It can adjust tone, imitate style, expand paragraphs, compress content, and make the result appear more complete and mature.
These abilities are valuable, and I also use AI to help organize and revise my writing. But the central question of expression is not whether the language has become more fluent. It is: Does this passage still belong to me?
After AI revises a passage, I need to ask again: Is this what I genuinely want to say? Does this tone reflect where I actually stand? Has this position passed through my own understanding and judgment? If I strip away the rhetoric, is what I truly mean still present? If someone responds, misunderstands, or challenges it, am I willing to stand behind these words?
The core of Expression is making clear where voice, judgment, and responsibility belong.
A text must not only be generated; it must belong to a particular person. Here, “belonging” does not mean that every word must be typed by hand. It means that the experience, judgment, meaning, and responsibility behind the words remain grounded in the person who is speaking.
If AI merely helps me express a judgment I have already formed more clearly, it extends my expressive capacity. But if the position, voice, and meaning are primarily generated by AI while I merely publish the result and bear its real-world consequences, then the process by which that expression is formed has already been outsourced.
This is why the question “Does an article written by AI still belong to me?” cannot be answered simply by asking who typed the words. The deeper questions are: Who formed the judgment? Whose experience gave rise to it? Who gave it meaning? When these words enter the public sphere and begin to affect relationships with others, who is willing to stand behind them? Who bears the consequences?
Responsibility cannot be generated by AI, nor does it disappear simply because the process of judgment has been outsourced. The ultimate question of Expression is: How do I articulate a judgment that is genuinely my own?
In the age of AI, the practice of expression is not about rejecting AI. It is about refusing to let AI decide where the voice belongs. If a passage does not correspond to a position I genuinely understand and am willing to stand behind and take responsibility for, then no matter how polished or beautiful it is, it is not worth publishing under my name.
At this point, I need to add an important clarification.
Evaluation, Judgment, Decision, and Expression are not the whole of Cognitive OS, nor are they four fixed steps that can simply be completed in sequence. They are four central domains of practice—and four of the easiest places to begin in everyday life.
A complete Cognitive OS also organizes a much wider set of processes: what captures our attention; what is recognized as a problem; how we form an understanding of the world, ourselves, and other people; how goals, values, and norms enter judgment; how action is initiated, delegated, suspended, or refused; and how real-world consequences feed back into the system and force us to revise our models and methods of judgment.
Cognitive OS does not reside in any single module. It lies in how these functions are organized, how they constrain and call upon one another, and how the system as a whole updates over time. It is better understood as a recurring loop:
The world enters cognition. Cognition forms judgment. Judgment becomes action. Action changes reality, and its consequences feed back into cognition, reshaping what we notice and how we question, understand, and choose the next time.
Expression may reveal that my judgment is still unclear. The cost of a decision may force me to reconsider my ordering of values. Failed action may expose a flaw in my model of the world. New evidence may require the entire problem to be framed again.
Real cognition has never been a one-way pipeline. Nor do we acquire a Cognitive OS only after learning the term. Every person already lives within some implicit cognitive architecture. It may have been shaped by family, education, culture, and relationships, and it may now be continually reshaped by platforms, metrics, and AI.
Rebuilding Cognitive OS does not mean installing a ready-made cognitive template from scratch. It means entering a continuous process: making hidden structures visible, examining them, reconstructing them, and continually updating them in response to reality.
In this essay, I have begun with four of the most direct training domains. Evaluation helps us see what we are relying on. Judgment helps us form a position that has been examined and that we are willing to stand behind and take responsibility for. Decision helps us initiate action amid uncertainty and cost. Expression restores ownership of our voice, judgment, and responsibility.
Together, these four domains create an essential cognitive space between AI output and human action. Instead of merely clicking “confirm,” we ask again:
Do I understand? Have I examined the reasons? Have I seen the conflict of values? Is this the direction I genuinely choose? Am I willing to bear the consequences? Are these still the words I want to say?
When this space remains open, AI can become a powerful cognitive aid. When it disappears, a person can easily slip from being the subject of judgment and action into becoming a confirmation interface for systems.
This is my current understanding of Cognitive OS.
It is not designed to make us use AI less, but to keep us from losing ourselves in the process of using it. It is not designed to give us more answers, but to train us to turn information into understanding, understanding into judgment, judgment into action—and the consequences of action into a renewed understanding of ourselves and the world.
But a more concrete question remains:
Where, exactly, can Cognitive OS be trained? What kind of practice can make these capacities part of our everyday lives?
My answer begins with writing.
In the next essay, I want to explore:
Why Is Long-Form Writing One of the Most Complete Everyday Training Grounds for Cognitive OS?
Because in an age when answers are becoming easier to generate, what we most need to practice is not merely how to obtain an answer, but how to experience for ourselves the full process by which a question appears, unfolds, becomes a judgment, and finally enters public expression and our relationships with others.