HubSpot lead scoring told us who was engaged. Not who would pay.

The trick: store the reasoning trace as a property reps can read, not a log in CloudWatch.

The Skill Library Has a Discovery Ceiling. Most Hit It Around 50.

Your AI Workflow Doesn't Need Better Prompts. It Needs Less AI.

Prompting is discovery. Skills are repetition. Gates are how AI workflows become reliable.

Claude Code with Local LLMs and ANTHROPIC_BASE_URL: Ollama, LM Studio, llama.cpp, vLLM

Run Claude Code on a local LLM via ANTHROPIC_BASE_URL.

Voice AI in Production: From RunPod to Hosted Kubernetes

One pod serves one user at a time. Production serves thousands.