As Agentic AI Reshapes Data Architecture, Data Leaders Need A Direct Line To The CEO

Agentic AI is moving faster than the org charts built to support it. Enterprises that spent the last decade wiring together cognitive-era architectures, dashboards, batch pipelines, and read-path analytics are now discovering that those systems were never designed to operate in the write path, where autonomous agents make decisions and take action. The technology gap is real, but the more consequential gap is structural: the people who understand how data should come together often lack the authority, the staffing model, and the reporting line to do anything about it.

Tom Gomez is Managing Partner of Luminity Digital, an applied AI architecture firm working at the intersection of data intelligence, agentic AI, and enterprise systems. He brings more than 25 years of experience across Fortune 500 companies in financial services, healthcare, and manufacturing, with prior leadership roles at Accenture, Capgemini, and Merrill Lynch, and has served as adjunct faculty in health informatics at Florida International University. Gomez grounds his advisory work in empirical research, publishing systematic reviews of peer-reviewed papers to guide clients through the transition from cognitive-era architectures to agentic AI. He said that however the technology stack evolves, the next wave of enterprise AI will be decided by who owns the data.

"The data engineer of tomorrow needs to know data engineering, analytics, machine learning, and generative AI, and have walked in their shoes. You're not really a data engineer because the agents are doing the work for you," Gomez said. The architecture is changing faster than the org chart. As agentic AI moves directly into enterprise workflows, the rote pipeline work that once defined data engineering is increasingly automated. The engineers who remain need to operate across the full intelligence stack, not just the plumbing.

The infrastructure most enterprises built to get here was never designed for what comes next. CIOs have invested heavily in middleware and agent frameworks meant to accelerate AI deployment, but many of those tools were built for the cognitive era, for prototyping rather than production. The same gap shows up on the people side: the data engineer profile is shifting well beyond traditional batch processing, and the leaders responsible for that function are increasingly caught between overlapping mandates and reporting lines that haven't kept pace with what agentic AI actually requires.

• Pinocchio protocols: The problem is structural, not incidental. Gomez attributed many AI production failures to how agent frameworks mask the stateless nature of large language models, a design flaw that compounds at scale. "Agent frameworks lie to the LLM. That is why you get hallucinations, and nobody talks about this," he said. "It is a leaky abstraction. When these frameworks interact with a tool, they lack parsing accuracy and fail to report errors back to the model."

• Night shift no more: The data engineering role is already being rewritten. "The data engineer of yesterday was the guy managing legacy ETL on the graveyard shift. Nobody knew he existed until 10,000 records went into an exception file," Gomez said. "The data engineer of tomorrow relies on AI-infused, agentified data engineering. CDAOs want these modern engineers because they understand this transition."

• The authority gap: The firms that have pulled ahead on data and AI share a structural trait: they gave their data leaders direct access to the CEO, not a dotted line into IT. That tension between CIO and data leadership is showing up across industries as agentic AI forces org chart decisions that were easy to defer. Goldman Sachs and AllianceBernstein built their modeling capabilities with direct CEO access for years. Gomez said most organizations haven't caught up. "The people that need the autonomy around how data comes together don't have the hands and the authority to do it. Until an organization breaks up and gives that person the authority, the CIO is not going to let go," he said.

The organizational and architectural changes Gomez described point toward the same conclusion: simplify the stack, and be deliberate about where complexity lives. He suggested enterprises move away from rigid ETL pipelines toward intelligent orchestration capable of supporting agents across multiple systems. That simplification starts with rethinking the relationship between storage and the model. But a leaner stack does not resolve the accountability question. It sharpens it.

• Just the lake: Former Snowflake CEO Bob Muglia put it plainly: a data lake and an LLM is all you need. Gomez said the framing resonated. "He didn't even say lakehouse. He just said a lake. It makes complete sense," he said. "Anthropic is already executing on this. They just released 13 agents for financial services that will handle everything from KYC to compliance." For CDAOs, the shift in data architecture is already reshaping how they structure their roadmaps.

• The dotted line: As the stack compresses, Gomez said the internal battles over tooling and ownership will matter less. What replaces them is a harder standard: organizations that build on governed platforms will need to put their name on what the AI actually does. "Governance is not security. Just because you have governance doesn't mean you are secure," he said. "Ultimately, the internal battles will not matter anymore. The focus is going to shift entirely to assurance. It will come down to companies explicitly putting their corporate name on the AI's actions."

The decisions CIOs make now about data architecture will determine how much runway they have when agentic AI moves from experiment to operation. Analyst research bears that out: enterprises with successful AI programs invest up to four times more in data and analytics foundations than their peers. The organizations falling behind are not short on technology. They are short on the people and authority structures to use it well.

That is the mandate Gomez left on the table. The stack will compress, the frameworks will consolidate, and the tooling decisions that feel urgent today will be made by the platform providers. What will not be decided for CIOs is who owns the data, who has the authority to act on it, and whether the org chart reflects that reality. As organizations develop their agentic AI strategy, those structural questions will determine who is ready to move and who is not. "If you don't look at the world that way, you'll be fooled into doing things that won't last," he said.