Agentic AI Is Easy to Deploy, Hard to Govern: Dallas 2026

Agentic AI is not hard to deploy. Plenty of enterprises are doing it right now, at speed, with confidence. The hard part, the part that the loudest voices in the room rarely talk about, is deploying it in a way that actually holds up. That survives contact with messy data. That earns trust from a workforce that did not ask for it. That produces outcomes a CFO can point to in a quarterly review. That part is significantly harder. These themes took center stage in a panel discussion at CDO Vision Dallas 2026

The panel was moderated by Rajitha Munugala, Director of Enterprise Data and Analytics at Western Alliance Bank, whose structured questions drew out the specific tensions each leader is navigating inside their own organisations. Joining her on stage were Tammye Garrett, Chief Data Officer at UT Southwestern Medical Center; Karthikeyan Ilangovan, Vice President of Data, Analytics and AI/ML at MODE Global; Sailesh Bharathwaaj Krishnamurthy, Head of the AI Center of Excellence at 7-Eleven; and Jagadeesh Tupakula, Head of Data and AI at UNFI.

The panel on agentic AI and the future of enterprise growth brought together Data and AI leaders from healthcare, logistics, retail, and financial services. What emerged was less a showcase of AI ambition and more a frank audit of what it actually takes to make agentic AI deliver value, and what gets in the way.

Before any agent can be trusted to act autonomously, the data it acts on must be trusted. That was the opening position from the panel, and it held firm throughout. Leaders from UNFI and Mode Global described enterprises still running on fragmented ERP systems, inconsistent master data, and siloed knowledge that lives in people's heads rather than in any architecture. The phrase "garbage in, garbage out" was not used lightly, it was used as a hard limit. Agentic AI, the room agreed, does not solve a data quality problem. It amplifies it. Enterprise data strategy must come before agent deployment, not after. The organisations investing in unifying and consolidating data at enterprise level are the ones positioned to generate trusted, quality outputs from their agents. Those skipping that step are building on unstable ground.

One of the most insightful exchanges of the afternoon came when the leaders from a national convenience retail chain and a top-tier third-party logistics provider converged on the same point from different directions: the word "agent" has become synonymous with automation in a way that is causing real strategic harm. When any process that can be scripted gets labelled an agentic AI initiative, organisations take on token costs, governance complexity, and measurement overhead that simple process automation would never require. The more precise framing that emerged was that agents are most valuable when the problem is too dynamic, too contextual, or too relationship-dependent for a script to handle. The logistics provider, for example, deployed an agent capable of negotiating loads, routes, and pricing in natural conversation, a task that traditional ML could not replicate. The retail AI lead takes a similar view: choose the right tool, not the fashionable one, and measure it against business outcomes, not sprint velocity.

Healthcare gave the panel its most grounded perspective on AI governance. UT Southwestern Medical Center operates across a health system, a research institute, and a medical school simultaneously. For their Chief Data Officer, agentic AI's primary value is efficiency, freeing clinical staff to work at the top of their licence rather than absorbing repetitive administrative load. But the condition is non-negotiable: human oversight stays in the loop, particularly for anything touching clinical decision support. Patient data is messy, secondary and tertiary uses are difficult to control, and the risk of harm is direct. The 10-80-10 framing that surfaced across the panel, humans set the intent, agents execute the middle 80%, humans verify the outcome, was not described as a limitation of current AI. It was described as the correct design for enterprise AI leadership at this moment.

The panel closed with one piece of advice from each leader. The consensus: understand what AI cannot do before deciding where to use it; build a strong data foundation and govern it responsibly; be strategic and do not mistake momentum for direction. As one panellist put it, governance and controls are not a checkpoint at the end of an AI project. They are part of every step of the implementation.

CDO Vision heads to San Francisco on 3 Jun 2026. If you lead Data or AI at an enterprise organisation, apply for your invitation at cdovision.aim.media. Seats are limited to 30+ senior leaders.

CDO Vision is the world's leading intimate gathering series for Senior AI, Data and Tech Leaders. Organised by AIM Media House since 2022, the 2026 Global Series spans 20 cities across 5 continents, bringing the world's most senior data and AI leaders together to share insights, strengthen networks, and shape the future of data-driven business. For more information, visit cdovision.aim.media or contact [email protected].