Can Agentic AI Bridge the Gap with Trusted Enterprise Data?

As artificial intelligence moves beyond providing information and recommendations, enterprise software is becoming more capable of reasoning, making decisions, and taking action across business processes.

It’s a noticeable shift that could fundamentally change how enterprises operate. But according to Manish Sood, CEO of Reltio, an SAP company, and co-author of Agentic Intelligence, it also exposes a challenge companies have wrestled with for decades: fragmented, inconsistent, and poorly connected data.

“There is tremendous enthusiasm around agentic AI right now, and for good reason,” said Sood. “Almost every company is experimenting with it in some form or the other. But there is a very significant gap between ambition and underlying readiness.”

Research conducted by Harvard Business Review Analytic Services underscores that readiness gap: 94% of organizations surveyed are exploring or implementing agentic AI, while only 15% believe their data foundation is truly ready.

“An AI demo can tolerate a lot of imperfections and gaps in the data,” said Sood. “An AI agent operating inside a real business context cannot. The moment we give software the ability to make decisions or to take action, the quality, timeliness, and context of underlying data becomes much more consequential.”

AI highlights legacy data challenges

Sood believes fragmentation, trust, and governance are three recurring barriers when it comes to AI adoption. Because data remains scattered across hundreds or even thousands of enterprise systems, governance models built for traditional enterprise systems must evolve for an environment in which AI agents can initiate actions.

While these are familiar challenges, the speed and autonomy AI introduces is much different.

Historically, humans have compensated for fragmented systems via tapping into institutional knowledge. Employees know which spreadsheet to check, which colleague to call, or which exception to make. AI agents do not inherently possess that organizational context.

“If a human has incomplete information, we can often recognize it, ask another question, or find someone who knows the history,” said Sood. “In comparison, an autonomous agent may simply act.”

The question for enterprises therefore shifts from whether they can build an AI agent to whether they can trust the environment in which that agent operates.

From digital filing cabinets to connected intelligence

Many organizations still operate around what Sood refers to as the “filing cabinet” model. When businesses digitized, individual functions created their own systems for marketing, finance, sales, supply chain, service, and human resources. Moving those systems to the cloud did not necessarily eliminate the underlying silos.

“The intelligence age requires us to move from storing information to connecting that information,” he said.

This connected foundation supports what Sood and his co-author describe as “co-agency,” a new relationship between people and intelligent machines.

Historically, enterprise technology followed a relatively simple model: humans made decisions and machines executed them. AI changes that equation. Machines can contribute speed, scale, pattern recognition, and continuous operation, while people provide judgment, experience, accountability, and an understanding of nuance.

“If the human sees one version of the customer and the agent sees another, you don’t have co-agency; you have conflict,” said Sood.

Turning insight into action

These implications extend beyond technology architecture as agentic intelligence requires companies to rethink strategy, investment, workflows, governance, and the relationship between people and machines. It’s a notable shift that moves from insight to action.

“Insight by itself does not create economic value. Action does,” Sood shared.

For example, predicting that a piece of industrial equipment is likely to fail can provide valuable information. But an intelligent system capable of coordinating maintenance, locating the necessary parts, and scheduling service before the failure occurs can potentially transform that insight directly into a business outcome. In order for this to happen, however, trusted information needs to be available when and where decisions happen.

According to Sood, four responsibilities are necessary for organizations preparing for this environment:

1. Connect critical enterprise data.
2. Contextualize it with relationships, history, policies, and interactions.
3. Operationalize that trusted context at the point of decision.
4. Govern how AI agents act.

“Define where an agent can act, where a human needs to intervene, how decisions are traced, and how systems learn from mistakes,” Sood explained. “The goal isn’t simply to make more data available to AI. The goal is to make the enterprise understandable through AI.”

Building for the intelligence age

As AI models continue to improve, Sood believes competitive advantage will increasingly come from proprietary data, business context, workflows, processes, and feedback mechanisms surrounding them.

Trust will also become more consequential as organizations delegate greater authority to AI. As a result, companies will need confidence not only in what an agent knows, but where its information came from, how current it is, and what boundaries govern its actions.

“The companies that win won’t simply be the companies with the most sophisticated AI,” said Sood. “They will be the companies that create environments in which AI can understand the business, operate with trusted context, and act responsibly.