Integrating Deep Research into Harvey

This post details the rapid integration of OpenAI's Deep Research API into Harvey, achieved in under 12 hours. It highlights the role of Harvey's AI-native architecture, specifically the Workflow Engine, in enabling this speed. The Workflow Engine's composable AI building blocks and 'thinking states' facilitated the seamless integration of Deep Research's reasoning and actions, allowing for user intervention and transparent evaluation of intermediate results. The post also emphasizes the successful incorporation of Deep Research's URL citations into Harvey's existing citation system for traceability and verification. Furthermore, it describes the use of AI-assisted development, including AI coding assistants and Streamlit, to overcome challenges in integrating a new, complex API with dynamic output types and product frameworks.