Blog
Thoughts on AI, developer tools, and building products.
Two Models Today, Meta-Harnesses Tomorrow: How I Code With Claude and Codex
I build with two coding agents at once: Claude as the orchestrator that plans and holds context, and Codex as a bounded implementer and adversarial reviewer. Across sessions, the two models catch almost completely different bugs, and that gap is the point.
Upgrading My AI-Powered Second Brain With Reflexive Search
Progressive disclosure optimized how Claude searches my Obsidian vault. Reflexive search optimized when. The second turned out to matter more.
Building a Personal AI Agent From Scratch: What I Learned About Agent Harnesses
I built a minimal AI agent framework that runs 24/7, talks to me through Telegram, and manages its own memory. The biggest lesson: the model is 10% of what makes an agent useful - the other 90% is the harness.
Why x402 Has My Attention
The convergence of stablecoins and AI agents is enabling a new category of machine-to-machine payments. x402 is the protocol making it real.
My OpenClaw Setup
How I set up OpenClaw as a personal AI assistant on a $20/month VPS — with automated newsletter triage, memory consolidation, and model routing to keep costs down.
My Personalized AI-Powered Second Brain
How I built a personalized knowledge management system using Claude Code and Obsidian — with custom slash commands, session hooks, semantic search, and a daily-to-weekly processing pipeline.