Long-form writing on agents, RAG, LLM infrastructure and AI safety. For a specific setup, see Guides; shorter notes go out in the newsletter. New essays go out by RSS, and every post is also available as markdown for agents: add .md to its URL.
2026
6 OctGolem: A Type-Safe AI Agent Framework for GoGolem is an open-source AI agent framework for Go with generics, zero dependencies, MCP support, and built-in retries, budgets, and human approval. Here is why I built it and how it works.10 SeptCan We Really Make AI Safe?A powerful model becomes a safety problem when it can act. Safer systems limit what a mistake can do.14 JulWhere Should the AI Actually Go?In traditional machine learning, framing the problem decided whether it was solvable. With LLM products, a similar decision has moved into the architecture: where should the AI sit in the process?7 MayThe Ship You Can't Dock: Architectural Debt in the AI EraHow architectural debt accumulates when the very ground underneath you is moving, and why building AI systems feels like sailing a ship that can't dock.29 AprScaling to 1,500 Concurrent Users: PgBouncer and Null PoolingHow I discovered that application-level pooling doesn't work for long-running AI requests—and what actually does.5 AprWhat Gemma 4 Actually Does DifferentlyGemma 4's 31B model is outscoring systems with 10x more parameters on Arena Elo. Here's the architectural reasoning behind why that's possible.15 FebInjection Is Not Influence: The Illusion of LLM MemoryAfter three years of building LLM applications, I've learned that LLM memory is fundamentally different from human memory.