Capital One develops AI platform with custom open-weight models to boost competitiveness

Capital One has developed a scalable multi-agent AI platform built around customized open-weight models, according to Kel Vanee, the bank's MVP of machine learning engineering. Vanee spoke at the VB Transform 2026 conference about how the bank's approach to AI differs from simply using off-the-shelf models. He emphasized that Capital One is focused on building its AI capabilities from the ground up.

This initiative is the result of years of strategic investment in data transformation and cloud technology. Vanee noted that these foundational steps were critical for the bank to adapt quickly to the current surge in artificial intelligence applications. With a solid technical background, Capital One has been able to make informed architectural choices that support its AI ambitions.

One key decision was to create a centralized, enterprise-wide AI platform that includes built-in governance. This approach allows the bank to leverage proprietary data to customize open-weight models effectively. The bank has also designed its own multi-agent orchestration harness, enabling a more flexible and efficient AI environment. This customization helps ensure that the AI tools are tailored specifically to Capital One's needs.

Moving forward, Capital One's emphasis on building rather than buying AI solutions may set a precedent in the financial sector. By investing in its own technology, the bank aims to stay competitive and innovative in a rapidly evolving landscape. The success of this approach could influence other institutions to consider similar strategies, highlighting the importance of customization in AI development.