We are building an operating system of AI agents that automate manual, repetitive work—starting with customer service. A natural starting point for building AI agents is to think about prompting large language models (LLMs). But what else needs to happen? The software engineering practices of taking agents into production and turning their output into automated work are complex and not well…