Structure Beats Magic
Timeline

Watch the work unfold

Not announcements — the working record. Every concept as it was named, every article as it was written, every pattern as it was recognised, together with the readings, sessions and second thoughts that prompted them. Most of it is small. That is the point: the thinking shows in the sequence, not in any one entry. Newest first.

A single line with five nodes, unevenly spaced and clustered toward the recent end - work accumulating in bursts.
Post

Your Annual Review Should Be a Query

Post

Did the AI Build Your App, or Is It Your App?

Post

Information Autonomy Needs an Information Model

Post

Four Eyes, One Source

Post

Why Your Data Model Needs a Decision Model

Post

The Smart If-Statement Belongs Inside a Decision Table

Post

Your Labels Are the Model

Post

One Term, One Meaning: What Aviation Knew About AI Before AI

Post

Karpathy Sketched It. I've Been Running It for Years.

Post

Karpathy's LLM Council Votes. It Should Look Things Up.

Post

The Guidebook Beats the Chatbot

Post

Two AIs Agreeing Is One Witness

Post

The Model Never Sees the Data

Post

The Brain That Learns From Everyone and Leaks Nothing

Post

Pseudonymised Is Still Personal

Post

My Second Brain Is Other People's Data Too

Post

Five Controls, Not One — A Data Firewall for AI

Concept

Gates

Post

The Search That Found Nothing

Post

A Successful Report Counts What Was Sent, Not What Arrived

Post

Eleven Out of 22,069

Post

Strict Rules Fail Loudly. Lenient Ones Fail Silently.

Concept

concepts — de SBM-conceptbibliotheek

Post

Your Stuff Is a Database (You Just Haven't Modelled It Yet)

Post

The Day Is Not the Topic

Reading

Outsource the task. Never the thinking.

Steven Bartlett (delegate, systematise, scale) read against Erik Scherder (the effort was never the inconvenience — it was the point).

Two books arguing across the table, and the only thing they accidentally agree on. Handing over the task is easy and mostly correct. But some of that work was not work — it was training. We removed friction on principle: most was waste, some was load-bearing, and almost nobody has been separating the two. The useful question is not *what can I hand over* but **which effort was doing the work on me?**

Post

The Negative and the Print

Post

Design Your Life on Your Own Data

Post

Digital Marie Kondo

Post

How I Restructured 700 Articles

Post

How to Build a Real Plan with AI (Not a Generic One)

Post

The Hundred-Line Photo Manager

Post

Verify Before You Destroy

Post

Local-First Versioning

Post

The Self That Improves Itself

Post

Your Data, Your Keys

Post

File Is Truth, the Database Is a Shadow

Post

Delete Is a Lie

Post

A Decision the AI Can't Find Is a Decision That Doesn't Exist

Post

One Filter to Rule Them All

Post

Define Only the Delta

Post

The Right Model for the Right Job (Or: What a Fast Model Is Actually For)

Post

One Foundation, Three Destinations

Post

A System That Documents Itself

Post

The Preference Layer

Post

The DuckDB Analysis Layer

Post

Organic Build, Periodic Integrate

Post

The Same Coordinate Means Something Different to You

Post

Satellites Over Sources

Post

Pay Once, Derive Forever — Storing AI's Interpretation So Rules Can Do the Rest

Post

Write Your Own Textbook — Personalized Learning Ebooks, and the First Working Piece

Post

The Map Already Knew

Post

The Interview Is the Escalation

Post

Let the Math Cluster, Let the AI Name

Post

The Gate That Wasn't There

Post

A Publish Layer Over a Local-First Source

Post

The Cost of a Copy

Post

Every Old Tool Is a Source — Pouring a Decade of Evernote Onto One Timeline

Post

Structure Your Workspace for AI — Why One Stable Root Beats Ten Scattered Ones