Pretext — Under the Hood exposes how to build structured LLM contexts, turning prompt construction into reusable, debuggable artifacts. Outcome engineers can adopt these patterns to make contexts testable and versionable, improving observability and reducing prompt drift (Principles 06 & 11).

Lat.md: Agent Lattice — a knowledge graph for your codebase, written in Markdown compresses codebase knowledge into interlinked Markdown files that agents can query and keep in sync with code. Use it to create a legible, agent-friendly Graph of product knowledge so retrieval is deterministic and easier to validate (Principle 11).

What if AI doesn’t need more RAM but better math? — How TurboQuant compresses the KV cache describes KV-cache compression techniques that slash memory demands for long-context LLM inference without losing accuracy. That shifts engineering trade-offs—longer contexts and persistent agent state become affordable, changing how you design Order and long-running agent workflows (Principle 12).

OpenYak — open-source desktop AI that runs any model and owns your filesystem ships a local-first desktop agent platform that manipulates files and automates workflows without cloud uploads. It’s a practical path to building private, high-capability agents, but it forces you to design Gate and Immune controls for local privilege and data access (Principles 15 & 14).

The Show Is Happening Right Now and Nothing Works recounts an AI-assisted live-music app failure that exposed debugging blindspots and breakdowns in human–assistant collaboration. Treat it as a case study: instrument live agent flows, add graceful fallbacks, and design team playbooks so agents don’t operate in single-player mode during live activities (Principles 03 & 06).