Scoping Agent Rules Like Routes, Not Like Prompts
Every session pays the token cost of every past edge case, whether or not it applies. Treating rules like routes instead of one long prompt is what fixed it.










Programmatic advertising has cleared machine-to-machine transactions for twenty years with no reliable way to prove who is on either end. The industry's workaround was to grade each transaction after it cleared, which catches fake traffic and misses a fake person. This walks through where the record actually breaks, how voice and likeness licensing works today, and what the draft ERC-8004 standard proposes instead.
Every session pays the token cost of every past edge case, whether or not it applies. Treating rules like routes instead of one long prompt is what fixed it.
Access to tools is the cheap half. Knowing when and how to use them is the part that costs context, which is why the knowledge lives in skills instead of the context window.
Three Claude surfaces run one engine, and Dispatch now routes work between them. What separates them is whether the setup shaping Claude's behavior is a file you own or state the app holds for you.
TypeScript / React / Three.js / GSAP / WebGL
If you want to go fast, go alone. If you want to go far, go together.
Design Engineering,Agents & MCP,LLM Orchestration,Durable Pipelines (Trigger.dev),Claude Opus 5,Cursor,Runway,Context.dev,Seedance 2.5,GPT Image 2,React / Next.js,TypeScript,Python,Creative Development,WebGL & Shaders (Three.js),Animation (GSAP),R3F,Tailwind,Postgres / Supabase,Cloudflare Workers,LLM Observability (Langfuse),UI / UX,
DesignEngineer.Agents,MCP,campaignsystems.
[KNOW MORE]
Studio. Fortune 500 campaign systems.