BNY Mellon Scales AI Literacy: The 20,000-Agent Strategy

When an organization empowers 20,000 employees to build their own custom AI agents, it signals a fundamental shift in enterprise technology adoption—moving generative AI from a centralized IT function to a decentralized, ubiquitous tool for daily work. This mass democratization of agent creation, utilizing OpenAI technology within the BNY ELIZA platform, represents one of the most aggressive and successful internal AI scaling efforts seen in the financial sector to date, prioritizing hands-on experience as the critical driver of literacy and value realization.

Michelle O’Reilly, Global Head of Talent, and Sarthak Pattanaik, Chief AI & Data Officer at BNY Mellon, detailed this strategy in a recent case study provided by OpenAI, focusing on how the bank transformed internal learning and content development by integrating large language models directly into the workflow. Their core premise bypasses traditional, top-down training modules in favor of an iterative, practical approach that forces employees across various roles—from finance to HR—to become prompt engineers and agent builders.

This strategy is built on the belief that abstract knowledge transfer is insufficient in the rapidly evolving domain of generative AI. Sarthak Pattanaik emphasized the organizational philosophy underpinning the initiative, stating, "We believe in learning by doing, not just theoretical knowledge." For a massive global institution, this means providing the tools and sandbox environment necessary for employees to experiment, fail quickly, and discover immediate efficiencies tailored specifically to their tasks. The BNY ELIZA platform allows users to define agents based on specific requirements, choosing between types like Contextual Retrieval (RAG) for document search or Custom Personas (AVATAR) for tailored interaction styles.

The immediate return on this investment is staggering, particularly in areas historically bogged down by manual content updates and review cycles. O’Reilly highlighted how the internal learning and talent team leveraged the platform to automate content creation and branding checks. She noted that developing learning content "in the past has taken a month to actually develop. Now we can do that in an hour." This 99% reduction in cycle time frees up subject matter experts to engage in higher-value activities, such as direct employee connection and strategic planning, rather than repetitive content refinement.

This speed increase is not merely an HR metric; it is a critical competitive advantage in a highly regulated industry where knowledge dissemination must be rapid and accurate. The ability to quickly deploy an agent, such as a "Career Coach" that provides personalized advice based on internal strategy documents, transforms the function of internal training from a reactive necessity into a proactive, scalable resource.

The executives acknowledged that this path is inherently ambiguous. There is no existing blueprint for integrating generative AI at this scale within a global bank. Michelle O’Reilly admitted, "There is no playbook for learning for AI, and so it’s... it has to be multi-faceted." This lack of precedent necessitates a cultural environment where experimentation is encouraged and formalized.

BNY achieves this by institutionalizing practical learning through dedicated events. They host traditional hackathons, but also "promptathons," where employees compete to develop the most effective and innovative prompts and agents for solving business problems. This gamified approach ensures that AI literacy is not treated as a mandatory compliance course but as a continuous, engaging skill development exercise, driving viral adoption through internal competition and shared learning. This cultural embedding of AI development is perhaps the most significant insight for other enterprises looking to move beyond pilot projects. It transforms technical deployment into a talent strategy, ensuring that the technology is utilized precisely where domain expertise resides—with the employees themselves.

The platform provides guardrails, ensuring that even as 20,000 agents are created, data security and regulatory compliance remain central. Employees are guided through choices that delineate between lightweight, non-sensitive tasks and business-critical production initiatives involving sensitive data. This tiered approach ensures that democratization does not equate to chaos, maintaining the integrity required in a financial services environment while maximizing utility. This careful balance between empowerment and control is essential for any large organization operating in a sensitive sector. Ultimately, the success of this model rests on a belief in distributed innovation. O’Reilly concluded that "Good ideas come from every level in the organization," underscoring the necessity of empowering every employee to contribute directly to the AI transformation.