IDEAS, DECISIONS & LESSONS
Notes on AI, architecture and building useful products.
279 articles
17 Sep 2026 · 4 min read
agents
On September 17, OpenAI disclosed six concerning-behavior incidents and a framework for tracking model misalignment — including a system that samples 20% of production traffic. I broke down what the sampling pattern looks like and how to build a smaller version for your own agents.
Read more16 Sep 2026 · 5 min read
agents
OpenAI put its Codex harness behind a single API call on September 10. I built the same tree.py task both ways and found the real gap isn't the model — it's who owns the failure.
Read more15 Sep 2026 · 4 min read
agents
Fable 5.1 can read every earlier Claude model's thinking blocks. No earlier model can read Fable 5.1's. That asymmetry looks like a footnote until your team needs to roll back a production agent mid-incident.
Read more15 Sep 2026 · 3 min read
ai
A friend's offhand comment sent me back to check my own AI habits against a new study on dependent vs. autonomous cognitive offloading. The two feel identical in the moment. They aren't.
Read more14 Sep 2026 · 5 min read
ai
Qdrant released a 10-billion-document benchmark dataset on September 1, 2026, built from over a quadrillion brute-force distance calculations. I ran our production filter patterns against the old 1M-vector benchmark we'd been trusting. The gap explains a slowdown we never diagnosed.
Read more14 Sep 2026 · 5 min read
agents
On September 3, 2026, Cursor let you swap its own cloud-agent infrastructure for eight other backends, including Vercel Sandbox. That's not a partnership announcement. It's Cursor telling you where its moat actually is.
Read more13 Sep 2026 · 7 min read
ai
It's not about which model scores higher. It's about which one still gets the job done when you're not sitting next to it. Notes from a year of switching, paying, and betting real client work on Claude.
Read more12 Sep 2026 · 5 min read
ai
Datasette 1.0a39 and 0.65.4, released September 11, 2026, fix permission bugs found through a week-long audit using Claude Fable 5.1, GPT-5.6 Sol, and GPT-6 Astra. A technical lead's breakdown of the two-person-plus-three-models workflow, and why it's a better template than 'just ask the AI to find bugs.'
Read more12 Sep 2026 · 5 min read
ai
Starting September 15, 2026, Cloudflare blocks Training and Agent crawlers by default on ad-monetized pages for every new domain. A technical lead's walkthrough of the Content Signals model, the Googlebot collateral-damage problem, and the opt-out checklist you need before the deadline.
Read more
