/ NotesNotes
- Agents in the Background: No LangGraph, No LangChain Required
How I actually run agents that watch folders and websites: a long-lived process, plain trigger scripts, cron — and no LangGraph or LangChain.
- Claude Code Auto Mode: The "Allow" Button Was Never Protecting You
Anthropic's classifier blocks 89% of dangerous actions; humans approve almost everything. Why Auto Mode by default matters — and what it doesn't replace.
- Mirroring Your Repo Is Not a Backup Plan
GitHub Actions and Pages went down and the vendor lock-in argument came back. I mirror every repo to three hosts — and still could not cut a release.
- Kimi K3: When an Open Model No Longer Means a Compromise
Kimi K3 matters less for its benchmark scores than for what it signals — the line between self-hosted and cloud models is starting to disappear.
- How I Split Work with Agentic Coding Tools: Architect, Reviewer, Agent
An AI model test became a shipped app. How I split work with coding agents: trust boundaries, written guardrails, and what I never delegate.
- "Cross-Platform" Means Three Separate Platforms: Shipping a Tauri 2 App
My Tauri 2 AppImage died with EGL_BAD_PARAMETER on recent Mesa: bundled graphics libraries. The fix lives in the release pipeline — and wasn't the only one.
- FAANG Experience Is No Longer an Elite Engineer Badge
Narrow roles, closed internal stacks, and golden handcuffs: why a Big Tech line on a résumé stopped being an automatic quality signal for hiring managers.