The Shape of Thought

Recently, it has become almost routine to hear people say that they have moved from one AI tool to another. Some speak of leaving OpenAI for Anthropic, often describing the shift in terms of workflow rather than raw capability.

There is something undeniably attractive about the emerging approach. A project folder, neatly organized. A collection of markdown files, each representing a unit of thought. An AI that moves through these files, connecting, refining, extending. It feels orderly. It feels intentional. It feels like progress.

I have felt that attraction myself. I have tried to build such systems, sometimes inspired by tools like Obsidian or environments like Visual Studio Code, where everything lives in plain text and structure. There is a certain satisfaction in seeing knowledge take shape, as if one were building a small architecture of understanding.

And yet, somewhere in that process, a question began to surface. If so much time is spent building the system, when does the thinking actually happen? At what point does preparation give way to the work itself?

The Engineer’s Dream

The appeal of these systems is not accidental. They reflect a long-standing habit of mind, especially among those who build systems for a living. When something is complex, the instinct is to structure it. When something is scattered, the instinct is to organize it.

Knowledge, in this view, becomes something that can be designed. It can be broken into parts, stored in files, arranged in folders, linked and versioned. Each piece has a place. Each idea has a container. Over time, the system grows, not through chaos, but through careful extension.

There is a kind of beauty in this. It offers a sense of control over what would otherwise feel overwhelming. It promises that nothing will be lost, that everything can be found, that understanding can be made durable.

In this light, the idea of a personal wiki or a second brain is not simply a tool. It is an expression of a deeper belief. That thinking itself can be stabilized, that knowledge can be made reliable through structure.

AI as Maintainer

With the arrival of AI, this vision has taken on new energy. What was once a static system can now evolve. Files are no longer just written. They are updated. Notes are no longer just stored. They are synthesized.

Much of the current excitement can be traced to ideas popularized by people like Andrej Karpathy, particularly his notion of an “LLM Wiki.” In this model, knowledge is continuously rewritten and refined by AI, turning a personal knowledge base into something closer to a living system.

The system begins to feel alive. It is no longer merely a repository, but a participant. It reads, it connects, it rewrites. It maintains coherence across a growing body of knowledge. The user is no longer the sole author, but part of an ongoing collaboration.

It is a compelling vision. A system that grows with you, that refines your thinking, that holds your knowledge in a form that is both accessible and evolving.

When Preparation Replaces Thought

And yet, something subtle can happen within this process. The act of maintaining the system begins to take on a life of its own. There is always another file to refine, another structure to improve, another connection to make.

The work becomes endless, but in a particular way. It is work that rarely demands a final position. It rarely forces a conclusion. It rarely asks you to commit.

Instead, it offers a kind of ongoing preparation. A sense that one is getting closer to clarity, without ever needing to fully arrive. The system grows, becomes more elegant, more comprehensive, more complete in appearance.

But the question remains. Has anything truly been thought through?

Writing as an Act of Thinking

Writing introduces a different dynamic. It does not allow for endless postponement. At some point, a sentence must be written. A choice must be made. A direction must be taken.

In writing, ambiguity cannot simply be stored. It must be faced. One cannot indefinitely refine the structure before beginning. The act of writing itself becomes the place where thinking unfolds.

This is why writing often feels difficult in a way that system-building does not. It requires exposure. It requires the willingness to be incomplete. It requires the acceptance that what is written may not fully capture what one intends.

And yet, it is precisely in that imperfection that something real emerges. Not a perfectly organized system, but a movement of thought.

The Limits of Structure

Structured systems favor clarity. They reward consistency. They depend on the idea that knowledge can be broken down into stable units and recombined as needed.

But human thought does not always follow this pattern. It moves through contradiction. It carries emotional weight. It shifts depending on context. It often resists clean categorization.

A system can store knowledge. It can even simulate coherence. But it cannot fully contain the movement of thought as it is lived.

Standing Between Two Worlds

None of this is to reject systems. They are useful. They allow us to manage scale, to retain information, to revisit what would otherwise be forgotten.

But they must remain in their place. As support, not as substitute.

A minimal system may be enough. Enough to hold what needs to be remembered. Enough to provide orientation. But not so much that maintaining the system becomes the primary activity.