Skip the AI ‘bake-off’ and build autonomous agents: Lessons from Intuit and Amex

As generative AI evolves, businesses are shifting gears from mere experimentation to practical implementation, venturing beyond basic chatbots to sophisticated autonomous agents. In a discussion during the VB Transform event, Ashok Srivastava, Senior Vice President and Chief Data Officer at Intuit, alongside Hillary Packer, Executive Vice President and Chief Technology Officer at American Express, shared insights on how both companies are leveraging agentic AI to enhance customer interactions, streamline internal processes, and improve overall business operations. At Intuit, the focus on intelligent agents transcends simple query responses; these systems are designed to execute tasks efficiently. For example, within TurboTax, these agents assist users in completing their tax returns 12% faster, with nearly half of users finishing in under an hour. The agents utilize data from various sources, including real-time and batch data, processed through Intuit’s internal frameworks. According to Srivastava, the goal is to create agents that are not only robust and scalable but also grounded in real-world applications. "We aim to ensure that the experiences we build truly accomplish tasks for the customer, always with their consent, which is essential for fostering trust," he remarked. The backbone of this initiative is Gen OS, Intuit’s proprietary generative AI operating system. Central to Gen OS is Gen Runtime, which functions similarly to a CPU, processing data, reasoning through it, and executing actions on behalf of users. The system is designed to minimize technical complexities, enabling developers to create agents without the need to repeatedly establish security measures. This capability is consistent across Intuit's various platforms, including TurboTax, QuickBooks, Mailchimp, and Credit Karma. On the other hand, American Express is building upon over 15 years of experience with traditional AI and a robust big data foundation to transition into agentic AI. The company is reimagining its strategy to leverage intelligent agents for enhancing internal workflows and crafting innovative customer experiences. A notable project is the APR agent, which enhances employee productivity by reviewing software pull requests and advising engineers on readiness for merging. This initiative aligns with Amex's broader strategy: prioritize internal applications, accelerate development, and utilize early successes to refine their infrastructure and governance frameworks. To facilitate rapid experimentation while ensuring security and compliance, Amex has introduced an “enablement layer” that allows for quick development while maintaining necessary oversight. Packer highlighted the importance of a control plane that integrates additional safeguards, stating, "Now, as we consider agentic AI, we have a solid framework to implement essential guardrails." The concept of modular “brains” is also pivotal at American Express. These frameworks require agents to consult specific “brains” before making decisions, ensuring adherence to brand values, privacy, security, and legal standards. This modular governance structure guarantees that agents remain accountable and aligned with organizational policies. For instance, a dining reservation agent through Resy, Amex’s restaurant booking service, must ensure it selects the appropriate restaurant according to user preferences while complying with brand and policy standards. Both leaders emphasized that scaling rapid development requires careful architectural planning. Intuit’s Gen OS allows numerous developers to build applications safely and consistently, ensuring shared infrastructure and risk management. Similarly, Amex's enablement layer is designed to facilitate the rapid development of AI agents while maintaining centralized governance. As they navigate this transformative landscape, both Intuit and American Express underscore the importance of speed coupled with strategic intent. Packer advised against waiting for a perfect solution, suggesting that it’s more beneficial to pursue a clear direction and iterate quickly. They also stressed that effective measurement must be embedded from the outset. Srivastava pointed out that tracking metrics such as cost, performance, and user impact is crucial for demonstrating value and accountability. "Gen OS includes built-in capabilities to monitor AI applications and assess both costs and returns," he explained. "I review these metrics quarterly with our CFO to ensure transparency in our AI investments." As frontrunners in adopting agentic AI, Intuit and American Express are not merely integrating new technologies but are redefining their operational models. Their comprehensive approach emphasizes building reliable agentic platforms, establishing governance frameworks, measuring outcomes, and advancing rapidly. As businesses aim to evolve from basic chatbot functionalities to fully autonomous operations, organizations that embrace agentic AI as a core discipline will be well-positioned to thrive in this competitive landscape.