Building the Context Layer
How context files give your AI agent the information it needs to produce more useful, specific output — and how to build them.
Product operator • 14 years • learning to build with AI
A practical record of learning what AI is actually useful for, where it falls short, and how it changes the work.
No hype cycle recap. Just useful experiments, working notes, and lessons from figuring this out in practice.
Working notes
How context files give your AI agent the information it needs to produce more useful, specific output — and how to build them.
A note on using personal context, visual references, Google Stitch, and DESIGN.md to build a website that feels less generic.
The repeated problems that pushed me to update a root Claude.md file, and the instructions I use to address them.
Toolkit
Prompts, guides, and working docs pulled out of the process and made reusable.
Agent instructions
My root Claude instructions file for anti-AI writing style, anti-sycophancy, critical thinking, identifying blindspot, coaching, and workflow defaults.
Prompt
A practical prompting reference with reusable frameworks, core techniques, advanced patterns, and a meta-prompt.
Projects
Shipped
Built by me, using an agentic development process. This personal website and publishing system powered by Next.js, MDX content, Vercel, Beehiiv newsletter capture, repo-level instructions, local skills, and a working log that keeps the build process visible.
Shipped
A shipped brand-system project built from an AI-led brand interview, visual-direction exploration, a homepage concept, and a coding-agent-ready design spec.
Experiment
An OpenClaw system for extracting and synthesizing useful ideas from YouTube, podcasts, and articles into structured notes.
Paused
A personal operating dashboard for AI learning logs, sports calendar and recipe-book.