Personal blog. Thoughts on agents, local LLMs, simulations, and whatever else comes to mind.https://dehydratedwater.dev/https://dehydratedwater.dev/blog/open-agent-compiler-design-choices/https://dehydratedwater.dev/blog/open-agent-compiler-design-choices/Design choices behind a Python-first compiler that produces OpenCode agents: deny-all-by-default permissions, bash as the universal tool surface, multi-model switching, and MCPs wired through a single build step.Sat, 18 Apr 2026 00:00:00 GMTagentshttps://dehydratedwater.dev/blog/moe-rule-binding-hypothesis/https://dehydratedwater.dev/blog/moe-rule-binding-hypothesis/A four-Qwen tool-use audit: the dense 27B sits at 5.6% errors, all three routed MoEs cluster at 10-12% regardless of per-expert capacity, fine-tune target, or quantization. Routing costs a roughly fixed reliability budget; post-training redistributes where it gets spent.Fri, 17 Apr 2026 00:00:00 GMTmodelshttps://dehydratedwater.dev/blog/qwen35-4x3090-optimal-agentic-inteligence/https://dehydratedwater.dev/blog/qwen35-4x3090-optimal-agentic-inteligence/Qwen3.6-35B wins on generation speed and Qwen3.5-122B on prefill per watt, but the dense Qwen3.5-27B makes about half the tool-call errors and recovers better when blocked — for my agent stack, reliability beats raw speed.Mon, 13 Apr 2026 00:00:00 GMTagents