Tutorials & Code

This section is for hands-on material that goes beyond a single technique: end-to-end builds that combine several of the deep-dive pages into one real, production-shaped project, rather than isolating one variable per recipe the way most pages on this site do.

Available now

• [x] Build a Production-Shaped Support Agent — real tool design, Claude's real memory tool, and risk-tiered guardrails combined into one support agent, tested against 6 realistic scenarios (including an embedded prompt injection) under ungated vs. tiered conditions — with a real, honest finding about the difference between "the model happened to resist" and "the architecture guarantees it."

Planned

• [ ] A multi-agent research assistant combining subagent division of labor with a real, deterministic verifier
• [ ] A durable, resumable coding agent with real crash recovery and idempotent tool replay