Type to search across all signals
No signals found
21 signals
Claude Fable 5 parsed every Claude Code transcript on the two machines I used in July 2026 and repriced 77,592 assistant turns at API list rates. The fleet ran the month for $6,648 in API-equivalent spend. The same tokens on the frontier model alone would have billed $15,482.
I pointed my Ralph Loop at its own script to ship a public demo. The useful part was not the recursion — it was every decision I made while the loop made the git log lie, burned $4 on an impossible spec, and taught me which corner I should have cut instead.
My first overnight Claude Code loop produced real commits and still failed: it burned quota faster than doing the work by hand. The real leak was the 68K-token runtime tax hiding before my prompt.
How Metis learned to speak — and why a human had to listen four times before it got the main word right.
Where the vault speaks back. The pipeline from captured knowledge to published content — and why the blank page problem is gone.
What the AI in this system actually does — and what it doesn't. Obsidian holds the knowledge. Lyra moves it.
Consequence: Capture became zero-friction via WhatsApp. The knowledge base became programmatically readable. The expression problem dissolved: instead of sitting down to write, I describe what happened. The vault routes it. The structure holds it. The AI layer expresses it.
How a WhatsApp message at 11pm becomes a structured knowledge base entry by morning — the raw input layer of the second brain.
The AI era didn't create new knowledge. It created an expression layer for people who already had it. This is for that breed.
What the vault actually looks like after three months of PARA and three weeks of automated routing. Real numbers from a real system.
Someone said my blog has LLM smell. They were right about the writing. They were wrong about where the knowledge came from.
How to generate blog cover images that actually reflect your content — using a three-tool pipeline built around visual metaphor, not stock photo logic.
Design Council is a Claude Skill that activates five distinct philosophical lenses on demand — Ive, Rams, Hadid, Ingels, Norman — each asking a different question about the same design.
How NanoClaw, the BMAD Method, and Claude Code Remote Control combine into a single workflow that goes from late-night insight to shipped code — regardless of what device you are on.
Open-source Claude Skills for design, architecture, and developer workflows.
AI generates plausible but wrong code in large legacy codebases — not randomly wrong, but wrong in ways that quietly erode conventions. Here's how to prevent it.
GPU-accelerated Rust terminal with local AI. Privacy-first, no cloud dependency.
NanoClaw, Kleos, and Nyx are three different ways to talk to the same AI. Here is why that architecture exists and how it came together.
Voice-first PWA and web dashboard for NanoClaw. React 19 + Vite 7.
AI doesn't replace engineers — it raises the floor. Stop trying to make humans consistent; make your tooling consistent instead.
Personal AI assistant running Claude in containers with isolated memory per context.