Logbook 2026-10-26

I’ve wanted a knowledge base that works for me for a long time. This is the story of how that idea finally started to become real.

For more than four years I’ve kept notes in Obsidian, following the Zettelkasten method. It earned its place: it’s where I keep and find my references, and it gave my memory some structure. But the method asks a lot, and over time the notes simply piled up. I never had a routine solid enough to review them, connect them, and actually take the time to draw insights from all that dormant material.

Two years ago I started using the Obsidian Web Clipper to make saving easier. I clipped a lot of links, mostly to gather references on whatever topics I was exploring at the time. Now and then I added a quote or a personal take, and I even started drafting a few article ideas, but I never took them any further. In the end, the system froze under its own accumulation.

Eighteen months ago, I tried bringing AI into it through Obsidian plugins, and none of them gave me anything convincing. I tested a local model, which at the time felt like the most accessible route for me. I asked it to link notes together, surface cross-cutting topics, write summaries. It simply wasn’t capable enough yet, and the results were disappointing.

The turn came from somewhere else. I’ve been a self-taught developer for a while, so AI coding tools came naturally to me, starting with Cursor on personal projects. After a break, hearing how Anthropic’s tooling had evolved, I restarted some stalled projects with Claude Code. I realised how much everything had changed: it would change for good the way I approach my projects, and push my ambitions past what used to limit me.

That is the real shift. With AI, all that dormant material can come alive: some of the work can be automated under my review, and the whole thing can adapt to the way I work.

I didn’t begin with code. I began with a conversation, asking an AI to help me plan without producing anything, and letting a brief take shape through questions and answers. The first build session kept things deliberately small: plain notes I can still read in Obsidian, a history of every change, no heavy infrastructure. By the evening, the first pieces were in place, though nothing real had gone through them yet.

Handing the work over felt like a real release. Tools can take on the time-consuming part, and I keep my attention for where I add the most value: judgement, asking the right questions, launching the right experiments. A bit like a researcher with tools that augment me, trying to crack the codes of tomorrow’s design systems.