Building and Managing an AI Agent Workforce

Moving Beyond 'Management' to Enablement

Alli K. Miller argues that the term 'managing' agents is a legacy mindset from 2015-era organizational structures. Instead of acting as a direct manager assigning granular tasks, founders should operate as architects who set high-level goals and infrastructure. The goal is to move from delegation to a 'liability' role, where the human provides final approval and critical thinking while the agentic workforce handles the execution and identifies new, proactive tasks.

The Proactive Agent Framework

Miller describes her own workforce, which consists of 34 agents organized into functions like education, operations, and marketing. She emphasizes the shift toward 'proactive' agents—systems that don't wait for explicit instructions but instead look at the business context (meeting transcripts, emails, Notion docs) to identify what needs to be done. A key prompt she uses is simply 'do smart things,' which relies on the agent's ability to reason across a vast, queryable knowledge base of the founder's life and business.

Designing for Ambition, Not Just Efficiency

One of the most distinct aspects of Miller's approach is hiring for roles that wouldn't exist in a human-only company. She highlights 'Phoebe,' a 'Chief Dreaming Officer' whose sole purpose is to challenge the founder and look for ways to 10x output. By treating agents as a product-led team rather than just an engineering task, founders can break through their own cognitive ceilings and avoid being the bottleneck in their own growth.

Implementation and Iteration

Building an effective agentic workforce is an iterative process. Miller suggests starting with a single agent, then moving to two agents working together, and eventually scaling to a full workforce. She stresses the importance of an 'AI watchdog'—using agents to monitor for friction, duplicative work, or cross-functional gaps. This creates a feedback loop where the AI identifies its own missing context or tool access, allowing the founder to refine the system over time.

The Arbitrage of AI-Native Workflows

Both speakers agree that because the majority of AI users are not yet building complex, multi-agent workflows, there is significant arbitrage for those who do. By creating a 'multiplayer' environment where human teammates can interact with the AI workforce via tools like Slack, companies can achieve operational speeds that were previously impossible for small teams.