Aug 22, 2026

An experimental editor that swaps longform prompts for pseudocode. You keep declarative pseudocode in .hz files; saving generates real code, and from then on the file's diff becomes the prompt. Human intent persists as an artifact instead of being discarded

Why it matters: Its naming of agent fatigue, the grind of longform English instructions, struck a nerve. Promoting prompts to first-class artifacts is a concrete answer to the lost record of human intent

A Go tool that pipes Claude Code's clipped terminal output through a local LLM to turn it back into readable English. Fully local, no telemetry; replace the output via hooks or tail it non-invasively on the side

Why it matters: Fixing a model's output with a second LLM is a self-deprecating patch that landed, helped by the author's candid disclaimers about hallucination and speed

An issue report: Codex CLI over Amazon Bedrock cannot opt into GPT-5.6 Sol's explicit prompt caching, so cache-write tokens balloon and real costs jump, backed by production data from an agentic coding workload

Why it matters: One caching knob can move an agentic bill by an order of magnitude, and the same model bought through a different route behaves like a different product