My AI agent tried to do a single line edit and got blocked by its own governance rules. That's not overhead — that's the system working. Here's what real AI throughput looks like when you stop babysitting and start governing.
Writing
Architecture notes on .NET, cloud systems, and AI integration. Working code, measured trade-offs, and production lessons.
2026
I spent more hours building AIAgentMinder than I did on the 11 sprints where I used it to ship a real product. A quick look at what that means — and what a two-minute slash command told me about the state of my project.
How to route automated and real-time Claude requests through your Max subscription instead of paying per-call — with two production patterns: an n8n sub-workflow for async operations and an Express gateway for synchronous app integration.
AIAgentMinder started as a memory system for Claude Code with rules and planing mode. Then Anthropic shipped native memory. Then they shipped rules. Then they shipped plan mode. Here's what survived, what didn't, and what I learned about building tools on a platform that moves faster than you do.
AIAgentMinder v0.7.0 removes three of four hooks after Claude Code's native memory system absorbed 80% of the project's value. What that means for tool builders designing around AI coding assistants.
Why I ditched an agent-agnostic approach and doubled down on Claude Code for true session continuity and cross-platform hooks.
I asked Claude Opus 4.6 for an exhaustive review of AIAgentMinder — my governance and lifecycle framework for Claude Code. Here's what it found, what it praised, and what it told me to delete.