Notes on AI agents, evaluation, and the craft of building intelligent systems.
Why AI coding agents fail on complex tasks — and how Claude Code solves the four fundamental bottlenecks of memory, planning, execution, and verification through scaffolding and multi-agent architecture.