No Circuit Fits All

Most LLM agent debates ask the wrong question. Instead of "is this mechanism good," the right question is "what work is this agent doing." LLMs are not appliances — they are electricity. The appliance emerges from the circuit: constraint set defines what freedoms the model has and doesn't have; topology determines how information flows, stops, decays, and feeds back. Engineering comes last. A memory mechanism that improves a research assistant can destroy an identity-emergence agent. Rational action selection that fits a task solver can flatten a being whose most important action is sometimes restraint. There is no universal agent architecture — only architecture relative to specific work.