Executives say AI agents need oversight before they can scale

Large companies are pushing AI agents into real business work, but executives from Citi, Experian, Ford and Dynatrace said broad adoption depends on knowing what the systems are doing and retaining the ability to intervene. Their comments at Fortune’s Brainstorm Tech conference showed how companies in finance, credit reporting and autos are putting controls around tools that can carry out chains of tasks.

Laura Heisman, chief marketing officer of Dynatrace, said companies are asking whether AI outputs can be trusted, whether they are correct and whether a faulty system can be stopped. She said trust requires visibility into the technology’s actions.

AI agents can link multiple steps together, with each step often relying on the output of an AI model. Heisman and other panelists said that makes monitoring and governance central to enterprise use, especially in regulated or safety-sensitive businesses.

Citi builds a central route for AI agents

Nikhil Joshi, chief information officer for Citi’s markets division, said visibility and traceability are core requirements for the bank. Citi moves trillions of dollars daily across more than 100 countries, according to Fortune.

Joshi said Citi spent much of 2024 creating a centralized technology base for its applications and AI agents. He said that structure has made the company more comfortable putting agents into production.

Under Citi’s approach, Joshi said, agents are deployed through a single central framework. He said each agent must be registered, monitored, audited and governed through that process.

Joshi said a cautious approach to AI can support speed later by giving companies a consistent structure for release and oversight. He described being “AI conservative” as a positive position for companies trying to scale the technology.

Experian tracks ownership and permissions

Kathleen Peters, chief innovation officer at Experian, said the consumer credit reporting company has built a system to manage the agents it deploys. She said the company tracks where each agent came from, which employee created it and what data or task permissions it has.

Peters said that shared understanding across the company helps create the trust needed to expand agent use and work more quickly. Her comments aligned with the panel’s broader view that companies cannot scale AI agents without clear records of control and accountability.

Ford uses AI to test ideas faster

Sammy Omari, Ford Motor Co.’s executive director for advanced driver assist systems and in-vehicle infotainment, said the automaker is using AI to speed parts of vehicle development and test ideas earlier. He said new vehicles can take years to move from design to production, making faster experimentation valuable.

Omari said Ford’s guardrails include limiting how AI-generated code is used. He gave the example of designers and other non-engineering employees using AI-powered “vibecoding” tools to create code for potential car features.

That code can help Ford see how a feature might appear in a test version of a vehicle, Omari said. If the concept fails, the company can drop it sooner; if it succeeds, engineers rewrite the code from scratch before it reaches a car sold to customers.

Omari said AI may accelerate speed to market, while Ford’s final quality-assurance process before customer release has not necessarily changed.

This story draws on original reporting from Fortune.