Writing

Nineteen essays on systems that had to work.

Agent architecture, evaluation, retrieval quality and the controls regulated teams ask for. Each essay comes out of work that shipped or work that failed.

Nothing is written to fill a schedule, and nothing claims a result it cannot source. The essays cover human-in-the-loop queues, model routers, telemetry envelopes, eval gates in CI, typed capsules, cache-aware prompts, durable agents and RAG quality.

Next

• Verification-First Architecture
• HITL Queues: A Design Doc for Production Human-in-the-Loop
• The Telemetry Envelope: What Every Agent Step Must Emit
• Your Judge Needs an Eval Too: Validating LLM-as-Judge in Production
• Eval-Driven Development: Four Gates That Belong in CI
• The Durable Agent Pattern: Temporal Meets LLMs in Production
• Zod In, Zod Out: Every Agent Step Is a Typed Capsule
• Context Engineering for Regulated Workflows: Four New Patterns
• A Router Is a Workflow, Not a Conditional
• Cache-Aware Prompts: Treating Prompts Like Build Artifacts
• GDPval & Guardrails: Real-World Benchmarks Meet Safety by Design
• Own the Automation Loop with an AI Partner
• Enterprise Agentic Workflow Pattern Language — Continued Addenda
• A Pattern Language for Enterprise Agents — Addenda
• Why a Pattern Language for Enterprise Agents?
• The Quality Engineering Loop for RAG
• Toward Reliable GenAI Systems: Why Simplicity, Context and Idempotence Beat Orchestras of Agents
• Enterprise AI Projects Fail at Scale Because They Lack Execution Infrastructure—Not Better Prompts
• From API Wrapper to Agent OS: Reframing MCP’s Role in the 2025 AI Stack
• RAG Quality Guide
• Pattern Language