Self-modifying agents constrained by Rice’s theorem

The author examines fifteen foundational results from computation, control theory, jurisprudence, and enforcement that constrain the design of a self-modifying software agent. The essay argues that each mechanism in the agent framework corresponds to a specific theoretical limit or necessity result, such as Rice's theorem on undecidability or Goodhart's law on metrics. While not claiming complete sufficiency, the author outlines the structural safeguards required to evaluate, test, govern, and stop an autonomous self-improving system. Open article →