• Building Your Skill-stack To Harness Agentic AI Use Cases o Overview Claude Code Overlay anatomy with the general anatomy of harnesses The guardrail architecture, which includes hallucination detection, completeness validators, pre-execution plan review, and human-in-the-loop gates o Build your Skill-stack Choose your either your own set of domain skills or pre-templated ones I'll provide • E.g. Coding Skills, Medical Skills, Research Skills, Data Science Skills, etc. • How To Observe And Score Agent Ability o Each agent and each human is characterized by a function from expenditure to score Expenditure will generally be treated as money, but it can also be interpreted as time or tokens o Quantify the capabilities of AI systems in terms of human capabilities Human-grounded metrics that compare agent and human expenditure require both expenditures to be measured in the same units o Not everything needs an agent How much of your control flow you hand to inference time, and what it costs you • Increasing Your Observability And Building Metrics and Benchmarks (even when using Claude) o Collecting data on your task completion and AI's task completion/responses Protocols, Telemetry, Networking, Files, Storage, and Data Management o Creating And Modifying Benchmarks for Task-Completion Ability And Supervised Fine-tuning o Distilling and Fine-tuning your own General-Purpose Models Gradually Building Your Own Training/Fine-tuning Datasets For Building Your Own Models