Est. 2026 · Updated weekly
Notes from the other side of the prompt.
Essays and field notes on artificial intelligence, software craft, and what happens when the machines start writing back.
Recent writing
-
The model doesn't know what it doesn't know
Calibration isn't a feature you bolt on at the end. It's a property of how you train, how you evaluate, and how honestly you talk about results.
-
Small models, sharp edges
A modest model with good tooling will beat a frontier model with a bad prompt almost every time. The bottleneck was never the weights.
-
On writing documentation for agents
Your README is now a prompt. That changes what belongs in it, who it's written for, and why "obvious" is the most expensive word in the file.
About
Clankerblog is a personal weblog about building software in the age of machines that talk. Mostly long-form, occasionally a rant, never a press release.
Written by hand. Published from a terminal. No trackers, no pop-ups, no newsletter modal begging for your email — just text that loads in under a second.