Posts
-
The Waiter, the Strawberry, and the Dinner Table That Broke the Machine
When I was a kid, my brain worked like a small hallucinating LLM. Not because I was doing matrix multiplication in the kitchen. More because I was very good at...
-
Sieve: The Same Failure, Smaller
There’s a weird little failure mode in coding agents that doesn’t look dramatic at first. The agent runs a test. The test fails. The terminal returns a wall of output....
-
The Models Learned Escalation From Us
Frontier AI nuclear “wargames” are less a warning about rogue machines than a mirror of the strategic archive that trained them. There is an easy version of this story, and...
-
The Protocol You Like Is Going to Come Back in Style
There is a certain kind of software dream that always arrives dressed like innocence. A little chat box. A few names in a sidebar. Some circles turning green. A message...
-
We Used to Talk About Tor. Well We’ve Got LLM Agents
There was a time when internet privacy debates had a relatively stable shape. We talked about Tor, VPNs, encrypted email, browser fingerprinting, metadata retention, and traffic analysis. The underlying model...
-
The Owls Are Not What They Seem
First post. The blog is called Fire Walk With Middleware. That should tell you enough about the tone. Writing about software, LLMs, and whatever I’m currently breaking or building. No...
subscribe via Substack or via RSS