Building an internal agent: Code-driven vs LLM-driven workflows

The blog post discusses the ongoing development of an internal agent that utilizes both code-driven and LLM-driven workflows. It illustrates the challenges and successes encountered in integrating LLMs into software processes, highlighting the need for determinism and efficiency in handling GitHub pull requests through automated workflows. The author shares personal insights from their project at Imprint, emphasizing a careful approach to problem-solving while balancing human oversight and automation.