Creating a Source of Truth for Your AI

One canvas. Every agent. One place to publish.

I've been writing this newsletter for more than three years. I have not missed a week. And in all that time, I have never once finished an edition and thought, wow, that was easy. Length, tone, the line between insight and rant — I still wrestle with the balance every week. Ten thousand hours in, and the work is still the work.

What has changed is the rig around the writing. For most of those three years I was tab-juggling: a draft in one app, research in ChatGPT, a second opinion from Claude, a project tracker somewhere else, and Wisprflow open to dump a voice note before I lost the thought. Every handoff between tools was a small tax on quality. Earlier this year, I admitted I had over-automated the whole pipeline with agents, and it cost me time and probably some quality too (the receipts are here).

Today the rig is simple—one workspace. Every agent reads from it and writes back to it. Mine happens to be Notion; for a real subset of you, it'll be Obsidian. Either way, you draft, edit, score, and ship from one canvas. This is the setup I would have given myself two years ago.

// The Takeaway: Pick one workspace — Notion or Obsidian — connect every AI tool you use to it via MCP, and do the work inside that editor. Your job stops being moving context between tools and starts being directing the agents that already have it.

Pick your path: Notion MCP → or Obsidian MCP →

Time cost: ~30 min

Tier: Free Notion plan or any Obsidian install

Tools: Notion or Obsidian, ChatGPT or Claude, MCP

For most of us, the AI problem is no longer which model. The models are good enough. The problem is that our context — the half-finished drafts, the source links, the brand voice notes, the publishing calendar — lives in five different apps, and every AI you talk to starts from zero. You spend the first ten minutes of every chat re-explaining your own work.

The fix is not another tool. It is picking one tool to be the home, and pointing the others at it.

// The real shift: You don't need the smartest AI. You need the AI that already knows your stuff. The win isn't a model upgrade — it's giving every model you already pay for the same source of truth.

Three reasons I landed here, in order of importance.

First, it has a real database underneath. My newsletter is driven by a Content Calendar that includes status, edition type, subject lines, publication dates, and metrics. Every edition is a row. Every row is also a full page I can write inside. That structure is what makes the agents useful, because they have something to filter, sort, and update — not just a pile of docs.

Second, it embeds AI directly in the editor. I can open any draft, ask Claude (running inside Notion) to score it against my own rubric, ask it to fill the subject-line field, ask it to draft the LinkedIn post in the property below, and the changes land on the page—no copy, no paste, no losing my place.

Third — and this is the new part — Notion now treats outside agents as first-class citizens. Notion 3.5, shipped May 13, 2026, opened a developer platform that lets Claude, OpenAI Codex, and other agents plug in directly. TechCrunch covered it as Notion turning the workspace into a hub for AI agents, which is exactly the right framing.

Obsidian is the credible alternative, and for real readers it is the better answer. The argument is just different.

It is local-first. Your notes are plain Markdown files in a folder on your disk. Nothing leaves your machine unless you sync it yourself, which means no vendor outage, no policy change, and no quarterly pricing surprise can lock you out of your own work.

It speaks MCP. The Obsidian MCP server lets Claude, ChatGPT, and any other MCP-aware agent read and write your vault the same way they read and write Notion. The plumbing is identical. What changes is who owns the disk.

What it does not give you is a database with views and filters. If your work has structured fields — a content calendar, a CRM, a project tracker — that's the trade. Dataview gets you close, but it is not the same as a real schema. Obsidian rewards the writer; Notion rewards the operator.

Pick Obsidian if you want files on your laptop, full control, and Markdown that will outlive any single product. Pick Notion if your work is half structured data and you want every AI tool to filter and sort it.

Each step is verified against the current product this week.

The shape of the work is the same. The steps look like this:

Notion AI is not a deep research engine. For broad web research, I still send Claude or ChatGPT from within Notion and have them write the findings back to a page.

The MCP connection authenticates via OAuth, and tokens can expire — you will occasionally have to re-authorize. Plan for it.

Custom Agents are a Business or Enterprise feature and now consume paid credits. If you are on the Free plan, you still get the in-editor AI and MCP, which is most of the value. The agent layer is where you start paying.

Obsidian's trade is the inverse: no hosted collaboration, no real-time multi-editor sessions, and sync across devices is your problem to solve. Plugins can break on updates, so pin the versions that work and don't auto-update the day before the deadline.

Neither setup is the whole stack. The full list of tools I run alongside this rig lives on my tools page.

For three years, the AI conversation was about which model was smartest. That argument is mostly over. The next two years are about context — which tool has the discipline to be the place your work actually lives, so the models you already pay for can do their job.

If you wait for the perfect setup, you'll still be tab-juggling at the end of the year. Pick the workspace, connect two AI tools to it this afternoon, and move one repetitive task off your plate next week — a Custom Agent if you're on Notion, a script or a Zap if you're on Obsidian. That is the whole lesson.