City Shift Finance: Why are so many companies struggling to turn AI adoption into measurable business value?
Adrian Wolfberg: Many companies struggle to turn AI adoption into measurable business value because they begin in the wrong place.
They ask, “How can we use AI?” before asking, “What problem are we solving, what judgment does it require, and where would machine intelligence actually improve the work?”
Measurable value comes when leaders connect AI to a clear business problem, define where human judgment remains essential, and design the workflow so human and machine intelligence fit together.
If the problem is poorly framed, AI may make the organization faster, but not better. A second issue is that companies often confuse activity with value. They count pilots, automations, and productivity gains, but those do not automatically create business value. AI may accelerate workflows, reports, or analysis, yet still fail to improve decisions, customer outcomes, or strategic adaptability. The question is not whether AI was used. The question is whether it changed something important. There is also a deeper leadership issue. AI can accelerate outputs faster than organizations can build shared understanding.
That creates what I call a clarity gap. The answer arrives quickly, but the human understanding needed to interpret and act on that answer develops more slowly. When organizations move too quickly from AI output to operational action, they create the appearance of progress while increasing risk. AI also exposes weaknesses that were already there. If decision rights are unclear, data practices are weak, or coordination is poor, AI will not fix those problems, it will amplify them.
City Shift Finance: As AI agents move from support to execution, where should leaders draw the line?
Adrian Wolfberg: As AI agents move from support to execution, leaders should draw the line where action creates consequences that require human judgment, contextual awareness, ethical responsibility, or accountability. The issue is not whether AI can execute a task.
The issue is whether the organization understands the conditions under which execution should occur and who remains responsible when it acts. There is an important difference between delegation and abdication. Delegation means leaders have intentionally designed the task, boundaries, escalation points, failure modes, and consequences of error. Abdication occurs when leaders allow AI to act simply because it is fast or technically capable, without preserving the human responsibility needed to govern it. For routine, low-risk tasks, AI may execute with minimal human involvement.
But as ambiguity, stakes, or human impact increase, the line should move closer to human control. In those settings, AI may support, recommend, or monitor, but humans should retain final authority. The line is also not fixed. It changes as conditions change. A task that is safe under normal conditions may require human intervention when the environment shifts. The practical test is simple: if the organization cannot explain why the AI acted, cannot detect when the context has changed, or cannot assign responsibility for the outcome, the agent should not act independently.
Execution authority should be earned through clarity, tested reliability, clear boundaries, and accountability.
The most important point is that accountability cannot be outsourced to machine intelligence. AI can contribute to the production of knowledge, but it cannot own the consequences of action. The more AI influences operational decisions, the more important human responsibility becomes.
Accountability should remain human, even when AI influences the decision.
AI can generate recommendations, identify patterns, automate analysis, or initiate action within defined boundaries, but it cannot carry moral, organizational, or legal responsibility the way leaders and institutions can. When AI influences operational decisions, accountability belongs to the people and organizations that designed, approved, deployed, governed, and acted on its output. That responsibility may be distributed. Senior leaders own the conditions of adoption. Technical teams own design, testing, and monitoring. Operational leaders own how AI is used in workflows. Frontline users own judgment within their role.
But distributed responsibility cannot become an excuse for no one being responsible. This is one of the biggest risks in AI adoption. When something goes wrong, organizations may say, “The system recommended it,” or “The model produced that output.” But systems do not place themselves into workflows. Leaders do. People frame the problem, select the data, set the objectives, and define the authority the system receives. That is why accountability has to be designed before AI is deployed, not debated after failure.
City Shift Finance: Are leaders moving too fast with AI without the right governance in place?
Adrian Wolfberg: In many cases, yes, but the issue is not simply speed.
It is speed without sufficient understanding. Leaders face competitive pressure, investor pressure, employee experimentation, vendor promises, and fear of being left behind. The impulse to move quickly is understandable. The problem is that many organizations are moving faster in adoption than in governance, learning, and judgment. Governance is often treated as compliance, but that is too narrow. Good AI governance is not only about rules, privacy, security, or legal review.
It also has to address how decisions are framed, how outputs are interpreted, when humans intervene, how accountability is assigned, and how the organization learns when systems perform differently than expected. The risk is that leaders confuse having a policy with having real governance. A policy may define what people should do, but governance has to shape workflows, decision rights, escalation paths, training, and feedback loops. Without that, organizations risk over-trusting outputs, creating fragmented experimentation, building hidden dependencies, and weakening accountability.
The answer is not to slow everything down. It is to match the pace of AI adoption to the clarity and risk of the problem. Low risk use cases can move faster. High-stakes, ambiguous, or people-affecting decisions require more deliberate governance. The real question is not whether leaders are moving too fast. It is whether they know where speed helps and where it creates risk.
City Shift Finance: What separates effective AI workforce design from wasted investment?
Adrian Wolfberg: What separates effective AI workforce design from wasted investment is whether leaders understand the fit between the problem, the technology, the workflow, and the people.
Too many organizations begin by asking what work can be automated. The better question is what humans should remain responsible for. when AI can perform parts of the work. That distinction shapes workforce design and determines whether AI creates value or simply adds activity. Some work can be automated because it is routine and rule-based. Some can be augmented because AI improves speed, pattern recognition, forecasting, or analysis. But ambiguous, high-stakes, or people-affecting work still requires human judgment, interpretation, ethical awareness, and accountability.
The organizations that gain from AI recognize those differences and redesign roles around them. AI will also change the meaning of expertise. People may gain access to capabilities that once required years of experience, which can create value but also create risk if answers arrive before understanding. That is why organizations need judgment literacy alongside technical literacy. People must know how to interpret outputs, question assumptions, and recognize when conditions have changed.
The companies that gain from AI will start with clear problem framing, design human-AI collaboration intentionally, preserve accountability, and build feedback loops as conditions change. The ones that waste capital will often chase novelty, automate poorly understood processes, and measure success by activity instead of outcomes. In the end, the difference is leadership discipline: knowing where machine intelligence belongs, where human judgment remains essential, and how to keep the two aligned.
Former U.S. Intelligence Officer and Organizational Scientist
