The system is simple in concept: Obsidian is where the work gathers, agents help turn raw material into usable drafts, I edit heavily, and Ghost is where selected pieces become public.
The actual machinery is messier, which is the honest version of any useful system.
Obsidian is the operating base. It holds daily notes, book summaries, content candidates, profile drafts, research outputs, trackers, source packs, workflow notes, and the small bits of institutional memory that make the next run better than the last one. I do not treat it as a beautiful archive. I treat it as a working surface.
The basic path looks like this.
First, something enters the system. It might be a Readwise highlight, a saved article, a conversation, a research paper, a founder profile candidate, a company deep dive idea, or a draft angle from a proactive editorial brief.
Second, it gets turned into an artifact. That might be a book-summary note, a research explainer scaffold, a profile source pack, a company deep-dive folder, or a content-series plan. The artifact matters because it gives the work a shape. It stops being "interesting stuff" and becomes "this draft, with this source basis, this intended reader, and this next action."
Third, it gets reviewed. Some review is editorial: does this sound like me, or does it sound like a machine trying to impress a committee? Some review is evidentiary: are the claims backed by actual sources? Some review is operational: did the draft get mirrored, indexed, tagged, and tracked correctly?
Fourth, if it clears the bar, it goes to Ghost as a draft. I like Ghost because it is clean enough to publish from and separate enough from the working system. Obsidian is where the thinking can be rough. Ghost is where the reader-facing artifact has to be coherent.
The tracker is the boring part and one of the most important parts. A content system without trackers becomes a pile of almost-finished work. The tracker tells me which series exist, where the plan lives, how many drafts are ready, whether Ghost drafts exist, which URLs were created, and what the next step is.
That last field, next step, is underrated. A lot of creative systems fail because they preserve material but not momentum. The note exists, the draft exists, the folder exists, but future-you has to reconstruct the state from scratch. A good next-action line is a gift to future-you. It says: start here.
Agents help most when the workflow is already explicit. They can extract text, create scaffolds, draft from source packs, run validators, check for repeated phrasing, update trackers, mirror folders to Obsidian, and create Ghost drafts. They are much worse when the task is just "make something good." The system works because the agents are operating inside a lane with constraints.
This is why the workflow has artifacts like source packs, claim ledgers, readiness reports, humanizer scores, and status files. They are not bureaucracy for its own sake. They are memory. They let the system ask: what was this based on, what changed, what passed, what failed, and what should happen next?
The biggest lesson is that publishing is not one workflow. It is a chain of handoffs.
Readwise to Obsidian is a handoff. Obsidian to draft is a handoff. Draft to validator is a handoff. Validator to editor is a handoff. Editor to Ghost is a handoff. Ghost back to tracker is a handoff. Every handoff can drop context. The system exists to make those drops visible.
I do not want a frictionless publishing machine. Friction is useful in the right places. It should be easy to capture and draft. It should be harder to publish something that has no source basis, no clear reader, no argument, or no reason to exist.
The workflow is not perfect. It has produced duplicate drafts, brittle titles, source-count edge cases, and posts that passed a mechanical check but still needed taste. That is normal. The point is not to remove judgment. The point is to make judgment easier to apply where it matters.
Obsidian gives me the working memory. Ghost gives me the public surface. Agents give me leverage across the repetitive middle. The system works when those three roles stay distinct.
The notebook is not a pile of notes. It is a publishing operating system: capture, synthesize, validate, publish, learn, and leave a trail good enough that the next pass starts smarter.
This is part 2 of 8 in Operating a Public Notebook.