This website isn't a blog. It's a working notebook.
A place to share observations, challenge assumptions and document what I'm learning.
Topics include:
A jar of jelly beans, an ox at a Plymouth livestock fair, and today's prediction markets all demonstrate the same trick: independent guesses, aggregated correctly, beat almost every individual guess. The mechanism is precise enough that AI systems now use it deliberately.
AI agents read the page and skip the visit, breaking the web's old ad-and-traffic bargain. Cloudflare's pay-per-crawl, Monetization Gateway and x402 protocol are the infrastructure being built to charge agents directly instead.
Models are shrinking, quantisation is improving, and consumer AI silicon is catching up fast enough that a growing share of everyday AI work can now run entirely on the device in front of you.
Optimisation has been quietly converging sport, design, film and music toward sameness for two decades. AI doesn't start that trend; it applies the same logic one layer deeper, straight to generation itself.
Wrappers, apps and services built on top of frontier models are being squeezed from both directions: enterprises building it themselves, foundation labs absorbing it into the model layer.
My aim is to publish ideas that are useful, practical and grounded in conversations with people doing the work every day.