Self-Programmed Execution for Language-Model Agents
The research introduces Self-Programmed Execution (SPE), where a language model generates its own orchestration program instead of relying on a fixed external policy. This is implemented using Spell, a Lisp dialect that allows programs to edit and re-evaluate themselves. Experiments show frontier models can learn Spell in context to solve complex agentic tasks, suggesting models can self-orchestrate without pre-set rules.