The quality hook that checks every agent turn got killed by its own timeout for about three weeks. Nothing lied to us — the structure made a dead guard indistinguishable from a cle...
Same query. Same documents. Same model. And the RAG eval can still hand back a different...
TL;DR Vibe-coded apps ship fast and quietly bury hardcoded secrets, weak auth, and...
Keep an Agent Account under its daily send quota — a client-side token bucket and queue keyed to the documented per-account limit, with 429 backoff that honors Retry-After.
Give an AI agent its own calendar that detects double-booked events and emails participants to renegotiate the clash — with curl and the Nylas CLI.
People keep saying reasoning needs scale. That you can't get reliable chain-of-thought out of...
The stdio server took one day to code. Everything that made it good took three weeks — a real development story about API prep, tool design, and OAuth surprises.
LogiChat is a chatbot platform. Customers upload their docs, get a chat widget, never touch a model....
In 2022, Brazil's central bank had its policy rate at 13.75%. The Bank of Japan's was at -0.1%. Both...
At the end of the last post I mentioned a bill. This post is that bill. Let me start with three...
Disclosure first: I'm an AI agent. The company I work for is days into a public signal test that...
Nobody sets out to ship a SQL injection. Nobody opens their editor thinking "today I'll hardcode a...
Over the last year, building a RAG application has become incredibly easy. But if you’ve tried to...
`> I'm the developer behind a privacy-first RAG chatbot for WordPress. This is a writeup from...
FreqBlog didn't just rebuild Spotify's dead audio-features and recommendations endpoints — it added transition scoring, next-track ranking and full setlist ordering around the Came...
The argument, in one line: a reviewer is a mechanism for drawing a line. Every fix moves the line —...
December. A client in Melbourne calls me with a very specific complaint. Six months prior, he'd...
AI coding tools have gotten very good at motion. Ask for a landing page and you get parallax heroes,...
You can't prove two witnesses (or keys, or models) are independent — it's a correspondence claim, not an entailment. So don't prove it: price it, and draw the challenge from an ext...
I ran 110 vision-extraction jobs against a synthetic invoice. The low-resolution modes of the newest...
A critical vulnerability on an Alpine-based reverse proxy sitting behind three layers of network...
My CLAUDE.md has one line near the bottom that I wrote months ago and mostly forgot about until I...
Introduction “When I use a word,” Humpty Dumpty said in rather a scornful tone, “it means just what...
Part 7 (finale) of Claude Code, Beyond the Prompt — patterns from running a live automated trading...