Getting Humans Out of the Way: How to Work with Teams of Agents

This episode of MLOps.community explores the evolving role of humans in software development as AI agents become more autonomous and capable. The host and guest discuss strategies for managing teams of AI agents, emphasizing the importance of designing robust validation processes—such as automated feature walkthrough docs with cropped, annotated screenshots and regression testing via Playwright scripts—to reduce human oversight. A core theme is teaching agents to 'manage up' by proactively reporting on their progress, identifying challenges, and asking the right questions, which shifts the human role from code reviewer to system architect and validator. The conversation also covers how to optimize agent workflows through custom lint rules, automated tooling, and parallel execution with conflict resolution handled by agents. The guest shares practical insights from using Broomy, an open-source tool that enables parallel agent sessions with customizable commands for merging, PR creation, and status checks. They stress that the human's new job is not writing code but defining what 'good' means, identifying solvable problems, and continuously refining systems to make agents more effective—essentially becoming a 'mini R&D department' in real time. The episode concludes with reflections on how rapidly AI advancements are reshaping roles, making adaptability and the ability to spot newly solvable problems the most valuable human skills.