For years, enterprise AI conversations have centred on chatbots, search assistants, and tools that respond when asked, but that era is ending. A new class of AI system, one that reasons, plans, and takes autonomous action, is moving from the research lab into live production environments. For C-suite leaders, the question is no longer if AI will arrive in their organisations, but whether those organisations are ready for it.

In a recent episode of Tech Transformed, host Christina Stathopoulos, founder of Dare to Data, sat down with Cathal McCarthy, Chief Executive Officer of Kore.ai, and Dan Leiva, founder of CXamplify and author of Amplified, to lay out what this shift actually means in practice and why most enterprises are less prepared than they think.

Have a look at Artemis, the agent platform from Kore.ai, or book a demo.

From AI Pilot Projects to Production

Most large organisations have run AI pilots. Far fewer have moved those pilots into meaningful production at scale. McCarthy and Leiva argue that this gap is not primarily a technology problem. It is a governance and accountability problem.

Conversational AI systems, which are the kind that answer questions or generate text, operate within a relatively contained risk envelope. A poorly worded response can be corrected, and a hallucinated answer can be flagged. The stakes, whilst real, are manageable.

Agentic AI operates differently. These systems do not simply respond to prompts. They assess situations, make decisions, trigger actions, and in some cases instruct other AI agents or software systems to carry out tasks on their behalf. When something goes wrong in an agentic workflow, the consequences can cascade quickly, across processes, data, customer interactions, and operational outputs.

This is why the move from pilot to production represents a fundamentally different risk conversation. As McCarthy puts it, "technology is now a decision-making actor." That framing has significant implications for how enterprises structure ownership, oversight, and accountability around their AI deployments.

What Agentic AI Actually Means for Your Organisation

The term “agentic AI” is often used loosely, so it is important to clarify what it actually means. An agentic system can:

  • Break a complex goal down into sub-tasks without human prompting at each step.
  • Use tools, APIs, databases, and other software to execute those tasks.
  • Adapt its approach based on intermediate results.
  • Operate across extended time horizons without continuous human input.

This is meaningfully different from a large language model that generates a report when asked, or a copilot that suggests the next line of code. Agentic systems take initiative, which means it's both their value and their risk.

Leiva's book, Amplified, explores how organisations can harness this capability without losing control of it. The central argument is that autonomy is not a binary switch; it is a dial. Organisations need to be deliberate about where they set that dial across use cases, risk profiles, and stages of deployment maturity.

A Framework for Smarter AI Decisions

One of the most practical tools discussed in the episode is the three-class decision model. Rather than treating all AI decisions as equivalent, it asks leaders to classify decisions by consequence and reversibility.

The first class covers routine, low-stakes decisions where agentic systems can operate with high autonomy, like scheduling, data routing, and standard customer queries. The second class covers decisions with moderate consequences, where human review should be triggered before action is taken. The third class covers high-stakes decisions where human authority must remain the final step.

Are you enjoying the content so far?

Mapping AI deployments to this framework is the foundation of a defensible governance structure, one that can satisfy board scrutiny and regulatory requirements simultaneously. It also forces a critical question: who owns the decision about which class a given AI action falls into? That ownership question, the guests argue, is where most enterprise AI programmes currently have a blind spot.

The Leadership Imperative

With that said, the organisations that will benefit most from the agentic era are not necessarily those with the most sophisticated technology. As Leiva writes in Amplified, they are the ones who have thought most carefully about how to deploy that technology in a way that is accountable, adaptable, and aligned with how their people actually work.

Boards are already asking harder questions about AI risk. Leaders who can answer them confidently because they have built the governance frameworks and defined the accountability structures will hold a material advantage. For leaders ready to move beyond the pilot stage, McCarthy and Leiva offer grounded guidance. Listen for more insights, and if you have any questions, feel free to get in touch with them directly.

Connect with the guests:

Further reading: Amplified by Dan Leiva — available on Amazon

Have a look at Artemis, the agent platform from Kore.ai, or book a demo.

Takeaways

  • The shift from conversational to agentic AI
  • Enterprise AI governance and accountability
  • Operationalising AI at scale and risk management
  • Building trust and transparency in autonomous AI systems
  • Turning AI experimentation into measurable business outcomes