- Two git repos isn't env parity
Sync Claude Code config across machines with a second git repo and you still get no environment parity: cloud sessions never read ~/.claude. Plus four bad fixes.
Sep 16, 2026 · 14 min
- The 401 is the map: three OAuth primitives in my private MCP servers
MCP OAuth 2.1 explained by walking one real client connection through a production server: discovery metadata, dynamic client registration, and the PKCE grant.
Jul 16, 2026 · 11 min
- Tools vs Resources vs Prompts: the MCP primitive you're misusing
MCP primitives (Tools, Resources, Prompts) are not interchangeable. The misuse that bites: shipping read-only data as a Tool the model must choose to call.
Jul 13, 2026 · 14 min
- MCP transports, decided
MCP transport choice, decided: stdio for local, Streamable HTTP for remote, SSE deprecated. The ten-second rule plus the stdout gotcha that costs you an afternoon.
Jul 1, 2026 · 9 min
- No, MCP servers aren't dead (and here's the 2007 reason why)
MCP isn't dead. The 'MCP is dead' wave measures the wrong axis: update friction. MCPs auto-fetch the maintainer's latest; skills default to download-and-own.
Jun 1, 2026 · 14 min
- Three Commands, Four Documents: The AI Workflow That Actually Ships
The Plan-Execute-Verify workflow that replaced my 13-document nightmare. Three commands, complexity scoring, and the failure modes to watch for.
Jan 23, 2026 · 9 min
- AI Agent Workflow: 13 Documents for a Label Change
A satirical anti-guide to AI agent workflows. Learn how to fail spectacularly by treating AI agents like a 2022 dev team. Conway's Law meets subagent chaos.
Jan 10, 2026 · 8 min
- Gemini 3 Pro vs GPT 5.2: A Blind Test Reveals Code Red DNA
Blind A/B test: Gemini 3 Pro vs GPT 5.2. Same prompt, same input. Gemini produced 60 surgical lines; GPT produced 250 bloated lines. Training beats prompts.
Dec 14, 2025 · 6 min
- Your AI Agent Config is Technical Debt You Haven't Acknowledged Yet
Your AI agent configuration is technical debt. From slash commands to skills, our prompts haven't evolved with the models. Time to audit your config.
Nov 28, 2025 · 6 min
- Three AI Agents Walk Into a Bar (and Only One Actually Read the Menu)
Real-world comparison of Claude Code, Gemini, and Codex AI coding agents tested on writing style analysis. Claude scored 9.5/10, Codex 7.5/10, Gemini 7.5/10. Configuration matters as much as capability.
Nov 20, 2025 · 9 min