Semantically Aligned Sentence-Level Embeddings for Agent Autonomy and Natural Language Understanding