Notion is the truth. Claude is the tool.

I don't use AI to get clever answers. I use AI to get things done. I'm also sick to death of the term. What I like is that it works, for one reason only: everything I know, have decided and am working on lives in one place. That place is Notion, my Single Source of Truth. Claude is the tool I do most things with, if not everything. Claude reads Notion, acts on what it finds there and writes the result back. Or adds something else, if I want it to.

Information everywhere

A decision was made in a meeting. The background is in an email, and the discussion carried on in a Slack thread. Tasks ended up in somebody's head, and three weeks later nobody knows why things turned out the way they did.

It has always been like this, it's just more expensive and more tiresome now. And AI on its own isn't the solution. An AI assistant is never better than the context it gets. If it doesn't know what you decided on Tuesday, which customers you've already talked to or how your product is built, you get generic answers. And generic answers are worthless when you're building something specific. So no, a free version of ChatGPT won't solve this.

That's why the question of which AI or "which prompt" is smartest doesn't interest me. The question that matters is how I make sure the tool always has the right context, without me having to paste it in every time. Or remind it, for that matter.

One source, one tool

1. Notion is the source. Everything worth remembering ends up there: projects, decisions, meeting notes, tasks, contact lists, specs and ideas. Notion doesn't actually do anything. It knows things.
2. Claude is the tool. Claude is connected to Notion via MCP (Model Context Protocol) and can search, read, create and update pages and databases on its own. Claude is also connected to everything else I work in: the codebase, hosting, database, email, calendar and browser. Claude is where I work, both in regular Claude and in Claude Code when I'm developing.

If you like, you can think of it as a computer where Notion is the hard drive and Claude the operating system. The value lies in the connection between them. So I care less about which model wins a given benchmark and more about how well the tools work together. So how did I end up here?

It started with receipts in Evernote

About fifteen years ago I got fed up with all the piles of paper at home and started scanning every paper I had: receipts, tax returns and contracts. All the way down to the kids' drawings. I fed them into Evernote one by one, because the goal was to never have to search through a binder again.

That turned out to be my first single source of truth, even though I didn't call it that at the time. The next step, maybe six years ago, was Notion, where I could structure the information in databases with relations, statuses and views instead of just storing it. It took a while to get the hang of Notion, but then. Damn!

Claude was the logical next step, and the big difference. You can use something else, I just happen to like Claude. Above all, I like that I'm no longer just finding information. I can get something done with it. And add new stuff if I want to.

How the source is structured

I try to keep it simple, and this is roughly what the structure looks like:

• Projects. Every client engagement, product and larger initiative has its own page, and everything else points here.
• Tasks. Databases with status, priority, project and deadline. Ordo, for example, has its own database called "Ordo – Tasks".
• Meetings. I record almost all my meetings. Transcripts and summaries are filed under the right project.
• Knowledge. Specs, technical descriptions, decisions and background, the kind of thing that would otherwise only exist in my head.
• Lists. Contacts, press, leads and sources. Everything that needs following up.

The point is that I'm not the one keeping all this in order. Claude is. Claude files things, links them together and keeps the source up to date. A structure that requires daily discipline won't survive a stressful week. A structure the tool looks after for you will. When a project gets messy, I ask Notion's own AI to tidy it up.

How it works

Ordo turns booking emails into smart itineraries. You forward your booking confirmations, and Ordo sorts flights, hotels, car rentals, safaris, wine tastings or restaurant visits into one itinerary. I'm building Ordo entirely on my own, and that's only possible because of this way of working.

Notion holds the work order, Claude carries it out

The "Ordo – Tasks" database isn't a to-do list in the usual sense. It works as a queue of work orders that Claude takes on. By work orders I mean that behind every heading there's a small wealth of information backing it up.

A typical flow can look like this:

1. I add a task in Notion: a bug, a new feature or an improvement. Sometimes it's one line, sometimes half a spec. Sometimes I write it myself, sometimes it's the result of something I've worked out in Claude. But there's always more than a heading.
2. In Claude Code, I ask Claude to fetch the next highest-priority task from "Ordo – Tasks".
3. Claude reads the task and whatever background it needs in Notion, such as technical documentation and earlier decisions. Then Claude reads the code in the repo.
4. Claude proposes a solution, implements it and tests it.
5. When the task is done, Claude updates it in Notion with the status, what was done and any follow-up tasks.

So the source stays current without me having to do it myself. The next time anyone needs to understand why something was built a certain way, whether it's Claude or me, the answer is there. If nothing else, it's a pretty good changelog.

The technical documentation lives there too

Ordo's entire technical stack is described in Notion: frontend, database, hosting, email services, payments, maps and flight data. That may sound like overkill for a one-person project, but it's exactly the opposite, damn it. That documentation is precisely what lets Claude make sensible decisions without me having to explain the architecture every time. I've also written into Claude's settings that it must NEVER change anything in Notion without a clear OK from me.

The division of labour

My role has changed. I write absolutely no code. Never have, never will. I'm responsible for strategy, priorities and taste: what gets built, why, and how it should feel. Claude handles the implementation.

This isn't a shortcut around work or experience, quite the opposite. It only works because I have a fairly long background in digital, but zero coding skills. I have to understand what's being built to be able to say no, change course and see when something is wrong. But my time ends up in the right place. And if I don't get something, Claude can always explain.

Everything around the product

A product is more than code, and the same principle applies to everything around Ordo:

• PR and outreach. My press list of just over 35 contacts lives in Notion. Claude sees who I've contacted, what they write about and who's next, and then writes a personal draft.
• Content and SEO. The blog on ordo.nu is filled with content through a flow that largely runs itself. Ideas, priorities and status live in the source, and Claude writes, optimises and publishes. I still need to keep an eye on it, but it delivers.
• Launch and funding. Launch material, pitch deck and investor contacts are built on the same source. When Claude writes an investor update, it starts from what actually happened, not from what I remember doing.
• Design. Ordo has a fairly well thought-out design system by now, for the simple reason that the service must under no circumstances look like it was coded by an AI. Colours, typefaces, tone of voice. It's all in Notion.

Claude as a filter

The smartest thing I've done is to see Claude as a filter for two kinds of information:

1. What I already have. Emails, Slack threads, meeting notes, quotes, ideas and documents I've written. Here Claude helps me find, summarise and connect things I would otherwise have forgotten.
2. What comes in. New emails, new meetings and new information from outside. Here Claude helps me sort: which project does it belong to, what requires me to do something, and what can I ignore?

Both flows go through Claude and land in Notion. Claude is what I work in. Notion is what I work towards.

The downsides

There are things that get in the way. Nobody said it would be a walk in the park:

• You become dependent on two vendors. When everything lives in Notion and all work goes through Claude, you depend on both. I think that's a reasonable price for what I get back, but it's an active choice, not a given. Export regularly. It's still fewer places than any workplace today.
• Garbage in, garbage out, only faster. If the source is messy, Claude's work gets messy. If you have no source at all, Claude will change its mind. The tool amplifies what you already have, both order and disorder. So keep things in order. I am NOT good at keeping things in order.
• It doesn't suit everyone. I only use the Office suite when I have to send something to someone who doesn't work the way I do. For many organisations it's exactly the other way round, and a way of working built on everything living in Notion is hard to bring into a team that lives in Teams and SharePoint.
• I still have to think for myself. Claude can fetch the next task, but I decide whether it's the right task. Prioritisation can't be delegated.
• Privacy. Meeting recordings and customer data in a system an AI tool can read require that you know what you're doing, and that the people you record know about it.

If you want to get started

My advice is not to build the whole system at once. It won't work. Focus on getting started. Maybe like this:

1. Pick a source. It doesn't have to be Notion, but it has to be one. Maybe starting with a Word document or a Google Sheet feels closest to you. Do that. Step up when you're ready.
2. Start with one project. Put everything related to it there: tasks, meetings and decisions. Pick one you want to get back to; a hobby or something you dream about. Something that engages you.
3. Hook the damn things up. Connect Claude to the source via MCP (if there is one) and start with simple questions. "Summarise where the project stands." "Which tasks are stuck?" "What did we decide about X?" How do you connect them? Ask Claude in a chat!
4. Let Claude write back. This is where the real value appears. When the tool doesn't just read but also updates, the system starts maintaining itself. But it may only write back what you say yes to.
5. Expand when it hurts. Add new databases and flows when you notice something is missing, not before.
6. Archive after a few months. If you haven't been in a project for a long time, archive it. It adds up very quickly. Don't delete, archive.

Sooo...

I don't think the competitive advantage of the future is about having access to the best AI model, because everyone has that. It isn't about that tonight or this morning either. The advantage lies in creating the best context and the best connections between the parts: a well-kept Single Source of Truth and a tool that can both read it and build on it.

For me, Notion is the source and Claude is the tool. Together they are proof that one person with zero coding skills, but the right system, can build what used to take a whole team.