Agents@Work: Benjamin Cox (Rakuten on Building AI Agents at Scale)

Quick Overview

Benjamin Cox, VP of AI Strategy & Operations at Rakuten Advertising, explains that successfully building AI agents at scale requires a fundamental mindset shift from traditional data science development towards an action-first, human-in-the-loop approach, focusing on automating repetitive workflows and providing agents as co-pilots to augment human capabilities rather than replacing them entirely.

Key Points: Successful scaling of AI agents requires a mindset shift away from purely model-building towards workflow automation, treating agents as co-pilots to augment human workers. Rakuten Advertising serves large advertisers across retail, finance, technology, and apparel, connecting them with large publisher populations online. The core challenge in scaling AI agents is determining which tasks to delegate to AI versus which should remain human-driven to ensure continuous improvement and maintain high-quality output. Cox advocates for an action-first approach, using agents to automate repetitive, low-risk, high-impact tasks that humans dislike, rather than attempting full automation of complex processes immediately. The rapid evolution of LLMs (like GPT) means that relying on outdated AI toolsets or assuming perfect information for model training leads to failure; continuous learning and iteration are essential. Rakuten has built internal analyst models and training for business process mapping to guide teams on identifying high-value, automatable tasks. Successful adoption relies on demonstrating tangible value (like increased productivity or reduced manual work) early on, preventing the feeling of being left behind by AI advancements.

Context: This video features an interview between Daniel Vassilev, Founder & Co-CEO of Relevance AI, and Benjamin Cox, VP of AI Strategy & Operations at Rakuten Advertising. The discussion centers on the practical challenges and strategic approach Rakuten Advertising takes when building and scaling useful AI agents within a large, complex corporate environment, contrasting the traditional data science approach with a more iterative, agent-centric workflow strategy.