The New Priority of the Agentic Age: Trust, Not Capability

There has been a clear shift in enterprise AI conversations over the past year. No one is asking, “What is a digital worker?” anymore; everyone is asking, “What can I delegate to a digital worker, and to what extent?” As digital workers move beyond demo environments and become embedded in operations, the central issue is changing as well: the focus is no longer capability, but trust.

McKinsey’s State of AI Trust in 2026: Shifting to the Agentic Era report substantiates this shift with data. According to the study, which surveyed executives responsible for AI governance and risk management across approximately 500 organizations, 62% of organizations are using digital workers at least on a trial basis, while 23% have already begun scaling them in at least one area. In other words, the agentic transformation is no longer a prediction; it is already happening in the field. This data tells us that the next divide will not be between organizations that have digital workers and those that do not, but between those that can confidently delegate work to their digital workers and those that cannot.

At CBOT, we do not treat trust as a compliance requirement or a policy document that is signed and filed away; we see it as the engine of scale. In this article, we explore the question, “How is trust built in the agentic age?” by combining McKinsey’s 2026 findings with the lessons we have learned from the systems we have built across finance, aviation, retail, and the public sector since 2015.

Enjoy the read,

The Issue Is No Longer the “Right Answer,” but the “Right Action”

When digital worker projects are discussed within organizations, two statements often follow one another: “The pilot was highly successful” and “However, we are hesitant to deploy it in production.”

The McKinsey report clearly identifies the source of this hesitation. According to the report, in the age of digital workers, organizations must now be concerned not only with systems saying the wrong thing, but also with doing the wrong thing. Taking an unintended action, misusing tools, or operating beyond defined boundaries brings trust directly to the center of operations.

At first glance, this distinction may seem like a technical detail. In reality, it fundamentally changes the nature of the work. When a chatbot gives an incorrect answer, the result is often a communication error that can be corrected. When a digital worker takes the wrong action, however, a transaction may be initiated, a record may be altered, or a process that directly affects the customer may be triggered.

We are seeing the same transformation in the field. As the digital worker evolves from a tool that merely offers recommendations into a system that initiates work, retrieves data, and executes decision-making steps, trust is no longer an abstract expectation. It becomes an architecture that must be designed, monitored, and continuously improved. The “handoff to a human” mechanism we discussed in previous issues was one of the practical ways to create value. In this issue, the same mechanism appears as one of the fundamental elements of trust.

Maturity Is Rising, and Those Building Balanced Capabilities Are Standing Out

The good news in the report is that organizations’ responsible AI maturity score rose from 2.0 to 2.3 in just one year. Organizations are learning, and their investments are paying off. The most interesting finding, however, lies beneath the average: approximately one-third of organizations have reached an advanced level—3 or above—in strategy, governance, and agentic governance.

This picture can be read as a list of shortcomings, but we interpret it from the opposite perspective. Technical capabilities are developing rapidly, while governance capabilities are still being applied systematically by only a limited number of organizations. This creates a clear opportunity for early movers to stand out. The pattern is familiar: last year, MIT’s research found that only 5% of organizations generated significant value from GenAI projects, while McKinsey reported 6% in 2025. A similar group of frontrunners is now emerging in the area of trust. Three different measurements point to the same reality: the majority are experimenting, while a small group is working systematically and pulling ahead.

The Most Cost-Effective Accelerator of Trust: Clear Ownership

So, what sets that one-third apart?

This is where the report offers its most actionable finding: Organizations with a clearly defined and accountable owner for responsible AI achieve a maturity score of 2.6, while those without clear ownership remain at 1.8. The gap is created not by a technology investment, but by a single organizational decision.

We see the practical impact of this finding every day in the field. When questions such as “Who approves it, who monitors it, and who intervenes when there is a deviation?” remain unanswered, even the most capable solution cannot move beyond cautious use. Every team gets stuck on the same questions, and scaling slows down on its own, without anyone actively applying the brakes. When those questions are answered, the opposite happens: operations teams feel secure, and the solution begins to expand. In the previous issue, we referred to this as “sustainable ownership”; McKinsey’s latest data has now quantified that observation.

Security Concerns Are Not a Brake, but a Design Input

According to the report, approximately two-thirds of organizations identify security and risk concerns as the number-one issue preventing them from scaling agentic AI—ahead of both regulatory uncertainty and technical limitations. It is easy to view these concerns as an obstacle, but we see something else: they are a sign that organizations are taking agentic AI seriously. No one convenes a risk committee for a toy that is still at the demo stage.

The right answer is not to slow down, but to design trust from the outset. Based on what we have learned in the field, this design consists of four steps:

First, appoint an owner. Responsible AI should have a clearly identified owner within the organization. The difference between 2.6 and 1.8 in the report demonstrates the value of this step on its own.

Second, define decision boundaries. What decisions can the digital worker make, based on which data, and up to what limit? When boundaries are clearly defined, autonomy ceases to be a risk.

Third, design the human handoff. Which exception should be escalated, at what threshold, and to whom? Human oversight should not be an approval box added later, but a clearly defined step built into the workflow from the very beginning.

Fourth, measure outcomes and expand through iteration. Trust is not declared; it is accumulated. The scope of authority granted to a digital worker that produces measurable results within a limited area should be expanded step by step, based on data.

This framework is most relevant in sectors where risk and regulation are particularly intensive: banking, finance, insurance, aviation, retail, and the public sector. This is no coincidence; the places where trust must become part of the architecture are those where the cost of error is highest.

Trust Is the Scaling Engine of the Agentic Age

If we were to summarize the entire report in a single sentence, it would be this: In the agentic age, scale comes not from more capable models, but from an accountable trust architecture. McKinsey’s data points in the same direction: Organizations that prioritize governance achieve measurably better results. In other words, trust is not a cost item or a compliance obligation; it is a competitive advantage in itself.

This is also the picture we see in the field at CBOT. The organizations that make digital workers a reliable part of their operations are not the boldest ones, but those that design trust most effectively. Every digital worker with clear ownership, defined boundaries, a structured human handoff, and measurable outcomes paves the way for the next one. The question is not, “Do you trust your agent?” The question is: Have you built trust as a scalable system?