Skip Bake-Off “AI and build autonomous agents: Lessons with Intuit and Amex

As a generative maturation is matured, enterprises pass from experiments to implementation – crossing chatbots and copiots to the sphere of knowledgeable, autonomous agents. In an interview with VentureBeat’s Matt Marshall, Ashok Srivastava, SVP and Director Director Intitutand Hillary Packer, EVP and CTO w American Express On VB TransformIt was described in detail how their companies accept the Artificial Intelligence Agency to transform customer experiences, internal work flows and basic business operations.

From models to missions: the creation of knowledgeable agents

In Intuit, agents are not only answering questions – they are aimed at performing tasks. For example, in Turbotax, agents support clients complete taxes by 12% faster, and almost half end in less than an hour. These knowledgeable systems download data from many streams-in this real-time and bastard-Via intitut’s internal bus and lasting services. After processing, the agent analyzes information to make a decision and take action.

“In this way we are thinking about agents in the financial field,” said Srivastava. “We try to make sure that when building they are solid, scalable and actually anchored in reality. The experience we build are designed to do the work Down Client, With their permission. This is the key to building trust. “

These possibilities are possible thanks to Genos, a non -standard generative intitut operating system. The heart is Genruntime, which Srivastava compares to the processor: he receives data, justifications and determines the action that is then made for the end user. OS has been designed to eliminate technical complexity, so programmers do not have to re -invent risk security or safety layers every time they build an agent.

Among the Intut brands – from Turbotax and Quickbooks to Mailchimp and Credit Karma – Genos helps to create coherent, trusted experiences and ensure reliability, scalability and expansion in the case of apply.

Building an agency pile in AMEX: Trust, Control and Experimentation

For Packer and her team in Amex, the transition to Agentic AI is based on over 15 years of experience with time-honored AI and mature, tested by the battle infrastructure of enormous data sets. As the Genai Amex is accelerated, it transforms its strategy to focus on how knowledgeable agents can drive internal work flows and supply the next generation of customer experiences. For example, the company focuses on developing internal agents who raise employee efficiency, like the APR agent, which reviews the demands of the software and advises engineers whether the code is ready for merging. This project reflects the wider Amex approach: Start with internal apply cases, move quickly and apply early victories to improve the basic infrastructure, tools and management standards.

To support quick experiments, mighty security and policy enforcement, Amex has developed a “layer of switching on”, which allows for rapid development without devoting supervision. “And so now, when we think about Agentica, we have a nice control aircraft to connect these additional, additional handrails that we really have to have on the spot,” said Packer.

This system contains the concept of modular “brains” Amex – frames in which agents are required to consult with specific “brains” before taking action. These brains serve as modular management layers-shifting brand values, privacy, security and legal compliance-that every agent must engage during decision making. Each brain represents a specific set of principles for the domain, such as brand voice, privacy principles or legal restrictions and functions as a consultation body. Through decisions through this system of restrictions, agents remain responsible, adapted to the company’s standards and trustworthy users.

For example, the gastronomic reservation agent operating through Rezi, the AMEX restaurant reservation platform, must confirm that he chooses the appropriate restaurant at the right time, matching the user’s intentions, while observing the brands and the guidelines of the rules.

Architecture that enables speed and safety

Both AI leaders agreed that enabling rapid development on a scale requires a thoughtful architectural design. In the intitut, the creation of Genos allows hundreds of programmers for sheltered and consistent building. The platform ensures that each team can access the common infrastructure, joint security and flexibility of the model without duplicating work.

Amex adopted a similar approach with the switching layer. Designed around a uniform control plane, the layer allows teams to quickly develop agents based on AI, while enforcing centralized policies and handrails. Provides consistent implementation of the risk and management framework, while encouraging to speed. Developers can quickly implement experiments, and then evaluate and scale based on feedback and performance, all without prejudice to the brand’s trust.

Lessons in Agentic AI adoption

Both AI leaders emphasized the need to move quickly, but with intention. “Don’t wait for baking,” Packer advised. “It is better to choose a direction, introduce something in production and quickly and items than delay the perfect solution that can be outdated during the premiere.” They also emphasized that the measurement must be embedded from the very beginning. According to Srivastava, instrumentation is not something that can be stuck later – it must be an integral part of the stack. The cost of tracking, delay, accuracy and user influence are necessary to assess the value and maintain enormous -scale responsibility.

“You have to measure it. This is where Genos appears-there is a built-in ability that allows us to be an AI Applications instrument and track both costs and return,” said Srivastava. “I browse this quarter with our financial director. We go according to each case of using AI in the whole company, assessing exactly how much we spend and what value we get in return.”

Bright agents are another change of the company’s platform

Intuit and American Express belong to the leading enterprises receiving Agentic AI not only as a technological layer, but as a fresh operating model. Their approach focuses on building an agency platform, establishing management, measuring influence and rapid movement. As the company’s expectations evolved from the uncomplicated functionality of Chatbot to autonomous execution, organizations that treat the AI ​​agency as a first-class discipline with controls, observation and modular management-it will be best to conduct an agent race.

Editor’s attention: As thanks to our readers, we opened early registration of birds for VB Transform 2026-Zaledwie 200 USD. Here, AI’s ambition meets with operational reality and you will want to be in peace. Book your place now.