Every LLM harness released in the last year has handled this gracefully via delegation, structured memory systems, etc.