Workflows then agents: the practical approach to enterprise AI

The text highlights the importance of reliable LLM workflows in solving real business problems, rather than chasing autonomous agent swarms. It emphasizes that most organizations don't need agents talking to agents to deliver immediate ROI from generative AI, but instead require well-orchestrated LLM workflows. The author suggests focusing on the simplest solution possible and only increasing complexity when needed. They also introduce an open-sourced AI SDK for Apache Airflow, which provides a declarative way to incorporate LLMs into workflows with built-in support for structured outputs, conditional branching, and agent integration. The SDK follows Airflow's taskflow pattern with three key decorators: `@task.llm`, `@task.agent`, and `@task.llm_branch`. This approach maintains the familiarity of Airflow while adding powerful AI capabilities, allowing organizations to deliver immediate business value while establishing a solid foundation for more sophisticated workflows.