Decision, Responsibility, and the Design of Judgment
The Paradox of Capability
We can do more than ever. Yet deciding feels harder than before.
This is not a contradiction. It is a consequence.
As generative AI and autonomous agents take on more of what we once called "work," a strange inversion is taking place. Execution—once the bottleneck of productivity—is becoming abundant. Drafts appear in seconds. Analysis materializes on demand. Code writes itself, or nearly so. The cost of doing has collapsed.
But the cost of deciding has not. If anything, it has risen.
Not because decisions have become more complex in some abstract sense, but because the infrastructure that once supported human judgment is being quietly dismantled. The friction that forced us to pause, to weigh, to choose—that friction is disappearing. And with it, something essential is slipping away.
This essay is about what remains when execution becomes cheap: the capacity to judge. And it is about why that capacity—what some now call "High Agency"—is not a personality trait or a motivational stance, but a structural problem that organizations and systems must solve by design.
What High Agency Is Not
The term "High Agency" has entered the lexicon of technology leadership, often attributed to figures like Sam Altman, who has described it as a defining characteristic of people who shape the future. In popular usage, it tends to evoke a certain kind of person: self-starting, relentless, undeterred by obstacles.
https://www.youtube.com/Wpxv-8nG8ec?si=52Gek_g2F0gmAlN8
This framing is not wrong, but it is incomplete—and in its incompleteness, it obscures more than it reveals.
High Agency, interpreted as motivation or mindset, becomes indistinguishable from ambition. It becomes a quality to be hired for, praised, perhaps even trained. It becomes, in other words, a human resources concept.
But the challenge we face is not a shortage of motivated people. It is the erosion of the conditions under which motivation can translate into meaningful decision.
High Agency, properly understood, is not about wanting to decide. It is about being able to decide—and being accountable for that decision in a way that is not merely formal, but substantive.
This means:
The capacity to meaningfully decide, not merely approve or delegate, and to remain accountable for outcomes.
High Agency is revealed not when the path is clear, but when it is ambiguous. Not when best practices exist, but when they do not. Not when the right answer can be looked up, but when it must be constructed.
And this is precisely where AI systems, for all their power, cannot substitute for human judgment—because ambiguity is not a problem to be solved, but a condition to be navigated.
The Risk: Hollow Judgment
There is a failure mode emerging in organizations that adopt AI at scale. It is subtle, and therefore dangerous. It does not look like failure. It looks like efficiency.
The failure is this: decisions appear to be made by humans, responsibility appears to be assigned to humans, but judgment has quietly migrated elsewhere.
Call it hollow judgment.
In hollow judgment, a human reviews an AI-generated recommendation and clicks "approve." The recommendation is reasonable. The human has no specific objection. The decision is logged, attributed, compliant. From an audit perspective, everything is in order.
But something is missing. The human did not weigh alternatives that were never presented. Did not question assumptions that were never surfaced. Did not exercise judgment in any meaningful sense—only performed its appearance.
This is not a moral failure on the part of the individual. It is a design failure on the part of the system.
AI outputs are dangerous not because they are wrong, but because they are often good enough. Good enough to act on. Good enough to defend. Good enough to make the act of questioning feel like obstruction rather than diligence.
When outputs are consistently good enough, the incentive to engage deeply with them erodes. And when that incentive erodes across an organization, judgment itself becomes vestigial—present in name, absent in function.
The human remains in the loop. But the loop has become a ritual.
Three Layers of High Agency
If High Agency is to be preserved—not as an ideal, but as a functioning reality—it must be understood as a system property, not merely an individual trait.
High Agency operates across three layers. Each layer is necessary. None is sufficient alone. And if any layer fails, the whole structure collapses.
The Individual Layer
At the individual layer, High Agency means that humans retain final judgment over consequential decisions. This does not mean rejecting AI assistance. It means treating AI outputs as inputs—material to be evaluated, not conclusions to be ratified.
The individual who exercises High Agency does not ask, "Is this output acceptable?" but rather, "Is this the decision I would make, given what I know and what I am accountable for?"
This requires not only capability, but also the structural permission to exercise it. A skilled professional who is punished for deviating from algorithmic recommendations will, over time, stop deviating. The individual layer depends on the organizational layer.
The Organizational Layer
At the organizational layer, High Agency means that decision authority is intentionally designed, not accidentally inherited.
Most organizations did not design their decision structures. They accumulated them. Decisions flow along paths carved by hierarchy, habit, and software defaults. When AI enters this landscape, it inherits these paths—and often optimizes them in ways that further entrench existing patterns.
An organization that preserves High Agency does not simply add AI tools to existing workflows. It asks: Where should judgment reside? Who should be accountable? What information must reach the decision-maker, and in what form?
These are design questions. And they require design answers—not policy statements, but structural choices about how decisions are made and who makes them.
Critically, judgment must not be structurally blocked. If the organization's systems, incentives, or cultures make it difficult or costly to override AI recommendations, then the formal presence of human decision-makers is meaningless. The organization has preserved the appearance of agency while eliminating its substance.
The AI Systems Layer
At the systems layer, High Agency means that AI systems are designed to preserve meaningful human intervention.
This is not the same as requiring human approval. A system can require human approval at every step and still eliminate meaningful agency, if the human is given no real basis for judgment, no time to deliberate, and no viable alternative to acceptance.
Systems that preserve High Agency are designed with what might be called decision boundaries—explicit points at which human judgment is not only permitted but expected. These boundaries are not afterthoughts or compliance features. They are architectural commitments.
At a decision boundary, a human can challenge, override, or halt a process. Not because they must, but because the system is designed to make such intervention possible and legitimate.
When AI systems are designed without decision boundaries, they do not eliminate human agency by force. They eliminate it by erosion. Each small optimization, each reduction in friction, each automation of a formerly human step—taken individually, these are improvements. Taken together, they hollow out the structure within which judgment once operated.
The Design Choice: Lose or Preserve
Every system embodies a choice, whether its designers recognize it or not.
Some systems optimize away judgment. They treat human involvement as a cost to be minimized, a source of error to be corrected, a bottleneck to be eliminated. In these systems, the ideal state is full automation. Human review is a transitional phase, tolerated until the algorithm is good enough to proceed alone.
Other systems preserve judgment by design. They treat human involvement not as a cost, but as a feature—a source of accountability, legitimacy, and adaptability that cannot be replicated by optimization. In these systems, the question is not "How do we remove the human?" but "How do we ensure the human can actually decide?"
This is not a philosophical distinction. It is an architectural one.
The concepts of Decision Design and Decision Boundary emerge from this recognition. Decision Design is the practice of intentionally structuring how decisions are made, by whom, with what information, and under what constraints. Decision Boundary is the identification and preservation of points at which human judgment must be exercised—not as a formality, but as a substantive act.
These are not methodologies to be adopted. They are lenses through which to evaluate existing systems and design future ones. They ask: Is judgment being exercised here, or merely performed? Is accountability real, or merely assigned?
Organizations that fail to ask these questions will not notice as agency drains from their structures. They will have dashboards full of approvals and audit trails full of names. They will have, by every formal measure, human oversight.
They will not have High Agency.
Conclusion: The Question That Remains
This essay has argued that High Agency is not a trait to be cultivated, but a condition to be designed. That judgment, in an age of abundant execution, is the scarce resource. That the risk is not AI failure, but AI success—success so consistent that it makes human judgment feel redundant.
But arguments are easy. Design is hard.
The harder question is not whether to preserve High Agency, but how to know whether you have.
Consider your own organization, your own systems, your own role:
Where is judgment actually exercised today? Not formally assigned, but genuinely exercised—where a human weighs alternatives, accepts uncertainty, and commits to a course of action for which they are truly accountable?
And where is judgment merely performed? Where does a human appear in the workflow, review an output, click a button, and move on—without ever engaging the ambiguity that would require real decision?
The difference between these two states is not always visible. It does not show up in process diagrams or compliance reports. It shows up only when something goes wrong, and we discover that the human we thought was deciding was only approving.
High Agency is not a slogan. It is not a hiring criterion. It is not a cultural value to be posted on walls.
It is a design problem. And like all design problems, it can be solved well or poorly, intentionally or by accident.
The age of AI does not eliminate the need for human judgment. It clarifies, with uncomfortable precision, where that judgment actually exists—and where it has already disappeared.
The question is not whether you value agency. The question is whether your systems preserve it.