Notes from the practice — short, occasional writing on AI, privacy, and the work.https://optimacore.io/https://optimacore.io/blog/the-three-generations-of-rag/https://optimacore.io/blog/the-three-generations-of-rag/Retrieval-augmented generation has moved through three generations — a fixed pipeline, an agentic loop, and a parallel orchestration over many sources. Read them as replacements and you'll reach for the newest and skip the foundation. Each generation wraps the one beneath it; the vocabulary of the first is the substrate the rest stand on.Fri, 26 Jun 2026 00:00:00 GMThttps://optimacore.io/blog/harnessing-volatility-in-production/https://optimacore.io/blog/harnessing-volatility-in-production/Agents demo beautifully and then fall over in production — not because the model is weak, but because its output is volatile and the system around it was never built to govern that. Reliability is an architecture problem, a governance problem, and an evaluation problem before it is a model problem.Wed, 24 Jun 2026 00:00:00 GMThttps://optimacore.io/blog/the-right-tool-for-the-loop/https://optimacore.io/blog/the-right-tool-for-the-loop/Agents now have to touch real systems, and two patterns have emerged to let them — the Model Context Protocol and the humble command line. Read it as a contest and you'll pick wrong. The question was never which protocol wins. It's which loop you're in.Tue, 23 Jun 2026 00:00:00 GMThttps://optimacore.io/blog/illusion-of-the-ai-operating-layer/https://optimacore.io/blog/illusion-of-the-ai-operating-layer/Enterprise AI is converging on a horizontal operating layer — gateways, protocols, generic agent UIs. That convergence is the trap. The defensible value isn't the wiring everyone shares; it's the governed, high-stakes domain underneath it.Mon, 22 Jun 2026 00:00:00 GMT