CDO Vision Seattle | Agentic AI and the Future of Enterprise Growth

Agentic AI is no longer a concept being debated in research labs or strategy offsites. It is being built, deployed, and, in many cases, quietly struggling inside real organisations right now. The promise is significant: AI systems that do not just answer questions but take action, complete workflows, and operate with increasing independence. But promise and readiness are two different things, and the distance between them is where most enterprise AI initiatives are currently living.

The question facing leaders today is not whether agentic AI belongs in their organisations. It does. The question is whether their organisations are structurally, technically, and culturally ready to support what autonomous systems actually demand. That gap, between ambition and operational maturity, was the defining thread of a panel at CDO Vision Seattle, part of the CDO Vision AI World Series.

CDO Vision Seattle brings together senior data, technology, and AI leaders to move the conversation on artificial intelligence from theory into practice. The AI World Series, of which Seattle is a part, has become one of the most important forums for executives navigating real implementation challenges across industries. The panel, titled Agentic AI and the Future of Enterprise Growth, was moderated by Anusua Trivedi, Head of GenAI at Oracle Health. She was joined by Shelby Tallent, Head of AI Governance, Risk and Compliance at Alaska Airlines; Venkat Surapaneni, Senior Technical Director of Data Strategy, AI and Cloud at Avalara; and Gajendra Babu Thokala, Senior Engineering Leader at Apple. Together, they brought perspectives from aviation, tax technology, and consumer technology, industries where the stakes of getting agentic AI wrong range from operational disruption to serious regulatory consequence.

Three themes emerged that cut across every sector represented on stage.

Maturity Gap

One of the clearest illustrations of where organisations actually are came from an aviation leader on the panel. Productivity tooling deployed at enterprise scale, with proper policy controls and security guardrails in place, had reached wide adoption. The same organisation's developer environment told a different story, only a handful of use cases had made it to production, all of them limited, all of them requiring a human in the loop. The reason was not that the AI was incapable. It was that identity management, access controls, and role-based permissioning were not mature enough to support anything broader. "Until we can ensure that an agent doesn't have a large enough blast radius," Shelby Tallent said, "our successes have just been with generalised documentation agents, functional agents, unit testing agents."

The panel's consensus was that autonomy has to be earned. It cannot be declared. Venkat Surapaneni pushed this further by making the case for model portability, building AI workflows across multiple model providers with an abstraction layer in between, so that no single model going down can collapse an entire system. The foundational work, model gateways, registries, governance layers, authorisation frameworks, needs to exist before use cases are built on top of it, not after. "Single agents working autonomously is easier than multi-agent workflows," Surapaneni noted, "and the autonomy level goes up only when humans trust the output." That trust, the panel agreed, is built through continuous measurement, model drift monitoring, and incrementally improving accuracy, not through confidence alone.

Governance is the foundation.

The panel was direct about how most organisations approach governance: too late, too reactive, and too disconnected from the people actually building the systems. Shelby Tallent described her approach as getting left of design, being present during ideation and sprint planning, understanding the technical maturity of each team, and identifying risk before it gets baked into architecture. "Governance should not derail or delay AI development," she said. "It should help deploy AI systems responsibly so that outcomes do not create harm to the business."

Gajendra Babu Thokala brought a concrete example of what happens when that principle is ignored. A major compliance project at a large technology organisation took several years and significant financial investment to complete — not because the problem was technically complex, but because governance had never been embedded in the original design. "If you don't do the groundwork now," Thokala said, "you are going to pay for it." The panel extended this to explainability, the ability to trace exactly how a model arrived at an output, what data it used, what tools it called, and how many reasoning loops it went through. Without that transparency, governance exists on paper but not in practice. Tallent added a pointed challenge to organisations: governance professionals need to understand how the technology actually works, not just how to cite the regulation it might violate.

The workforce is dividing, and the divide is widening.

The final theme was the most human, and in some ways the most urgent. Agentic AI is not affecting all workers equally, and the panel was candid about where the pressure is falling hardest. Gajendra Babu Thokala identified a specific crisis emerging among early-career professionals: AI is compressing the experience ladder. Senior practitioners can use agentic tools to delegate junior-level work and evaluate the output confidently. But junior professionals cannot reach senior level without the experience that comes from doing that work themselves. "They expect you to be a senior," Thokala said, "but juniors cannot get to senior level unless they have experience."

Venkat Surapaneni named the organisational split that is already visible inside many companies, one side rapidly building fluency and competitive advantage with AI tools, the other side either resistant or simply not keeping pace. The response, the panel agreed, cannot just be a policy document. Shelby Tallent described building an AI literacy infrastructure: role-specific training, a prompt library, clear guidance on what people can and cannot do, and a deliberate listening tour to understand what employees are actually afraid of. "People just want to know where they can go and where they can't," she said. "When you don't have that laid out, that's where you get shadow AI and agent sprawl."

The panel at CDO Vision Seattle did not offer easy reassurance. What it offered was something more useful, an honest account of what operational readiness actually looks like from people who are building it under real pressure. Agentic AI is not arriving. It is already here. The organisations that will benefit most from it are not the ones moving fastest. They are the ones building the foundation that makes speed sustainable.