Writing
De-composer
How I got Cursor's cheap, capable Composer model running from Claude Code while preserving local shell, file, edit, and MCP tools.
Clawdroid
Mobile apps as a tool surface for AI agents on Linux. Waydroid plus an accessibility bridge so OpenClaw can drive Amazon, Uber, and Instacart without tripping CAPTCHAs.
Strata: Session Archaeology
AI agents lose context when conversations grow. Strata compacts session history into structured, verifiable fossils — not summaries. A 4-phase pipeline that extracts structure instead of summarizing prose.
Attention Was All We Needed
How AI is redefining what it means to pay attention, learn, and multitask. A personal exploration of cognitive viscosity, chunking, and the loading screens we need.
Thermostat
Can an LLM change its own temperature mid-response? An experiment in letting the model decide when to be creative vs. precise — and why it probably doesn't matter.
Steering Playbook
From career pivots to emergency planning — copy-paste prompt frameworks for investigations, resumes, travel, moving, and more.
The Wheel Spinner vs. The Orchestrator
Stop pedaling and start steering. The mental models that separate productive AI orchestrators from wheel spinners — and why 2026 is the year to level up.
Your friend asks you "how do I screenshot on my phone?" for the 47th time. Instead of answering, send them this link. They'll figure it out, feel ridiculous, and never bother you with basic questions again.
DJFA opens the LLM of your choice, types their exact stupid question, and hits Enter. This passive-aggressive action will train them to ask AI first next time.