If Codex can read files, edit files, search the repository, and draft articles, why open Obsidian at all?

The question is fair. My answer is that Obsidian is not there to replace AI or automatically compile knowledge. It is there to let a human see, review, and steer the knowledge system.

Obsidian is the human interface

AI is good at local tasks: organize a source, extract a concept, compare two options, draft a publication source.

But a knowledge base is not only for AI. Durable knowledge must remain readable, editable, navigable, and challengeable by a human. That is where Obsidian is useful.

It gives me a human control surface:

  • links between concepts
  • backlinks that show where a claim is used
  • full-text search across old material
  • navigation through topics and maps
  • direct editing of Markdown files

AI can generate structure. Obsidian helps me decide whether that structure still makes sense.

That is why I care about Obsidian’s basic model more than any “AI graph magic.” Obsidian’s own documentation treats internal links as a core way to connect notes, and backlinks help a human see where a note is referenced. In an AI knowledge base, that is a practical affordance: it lets the human inspect the structure produced by AI instead of only trusting a summary.

Obsidian is not a truth machine

Calling Obsidian a “second brain” can be misleading. It makes it sound as if putting material into a vault automatically creates knowledge.

It does not.

Obsidian is a strong local knowledge interface. Knowledge still requires sources, judgment, compilation, revision, and deletion. Without those processes, Obsidian is just a nicer folder.

In this system, Obsidian is the review and navigation layer. It does not decide what should be promoted from a project. It does not generate public articles by itself. It does not replace publication review. It lets a human inspect the system.

Codex and Obsidian have different jobs

Codex is closer to a knowledge compiler.

It can turn raw material into concepts, project retrospectives into topics, and source drafts into publication candidates. It is useful when the task is explicit.

Obsidian is closer to a knowledge cockpit.

It helps me see whether the graph is coherent, whether a concept is isolated, whether a topic has grown too broad, or whether a public draft lacks evidence.

The division looks like this:

Codex compiles
Obsidian reviews
Git records
site publishes

If I only use Codex, the knowledge base can become a pile of AI-generated files. If I only use Obsidian, the cost of organizing project knowledge is too high, and many lessons never get captured.

Local Markdown is the key

Obsidian works in this system because the underlying material is local Markdown.

That means the knowledge is not locked inside a SaaS product or trapped in a chat provider’s history. The same files can be read by Obsidian, Codex, search tools, Git, scripts, and the publication adapter.

The same knowledge can have different interfaces:

  • Humans read and revise it in Obsidian.
  • AI tools compile it through the filesystem.
  • Scripts check metadata and paths.
  • The site renders reviewed generated artifacts.

That openness matters more than any single tool feature.

When I actually open Obsidian

I do not open Obsidian for every task.

If I only need Codex to clean up a file, check a project fact, or generate a draft, the command line is enough.

Obsidian becomes useful when I want to:

  • see where a concept is referenced
  • judge whether a series is structurally complete
  • find isolated or duplicated pages
  • manually rewrite an important definition
  • browse the knowledge system as a human instead of asking AI to summarize it

Obsidian is not smarter than AI. It gives the human a slower, more inspectable surface for judgment.

In an AI knowledge system, humans still need a place to pause, look, and decide. That is Obsidian’s role.