Essays on AI architecture, governance, and systems that ship.https://langoni.me/https://langoni.me/writing/ai-agent-approval-fatigue-is-real/https://langoni.me/writing/ai-agent-approval-fatigue-is-real/If your coding agent asks for approval on every harmless action, people either turn it off or stop using the loop. The fix is not more prompts. It is a policy layer.Sat, 11 Apr 2026 00:00:00 GMThttps://langoni.me/writing/the-missing-layer-in-autonomous-agents-is-runtime-policy/https://langoni.me/writing/the-missing-layer-in-autonomous-agents-is-runtime-policy/Teams trying to productionize autonomous agents usually do not fail because the model is weak. They fail because there is no clean policy layer between the agent and the tools it can trigger.Sat, 11 Apr 2026 00:00:00 GMThttps://langoni.me/writing/why-im-building-an-ai-operating-system/https://langoni.me/writing/why-im-building-an-ai-operating-system/Most AI tools are isolated workflows. I wanted to understand what happens when intent, execution, governance, and learning belong to the same system.Thu, 02 Apr 2026 00:00:00 GMThttps://langoni.me/writing/ai-code-division-of-labour/https://langoni.me/writing/ai-code-division-of-labour/Reliable agentic systems get faster and more truthful when models stop doing deterministic work and start handling only ambiguity.Fri, 20 Mar 2026 00:00:00 GMThttps://langoni.me/writing/zero-fabrication-by-architecture/https://langoni.me/writing/zero-fabrication-by-architecture/If truth matters, the answer is not a stronger prompt. It is removing the model from any path where it can invent facts.Sun, 08 Mar 2026 00:00:00 GMT