AI no longer supports judgment alone. Increasingly, it participates in judgment.
AI now drafts the memo, screens the candidate, and flags the suspicious transaction. In most organizations it has stopped being a project and become part of the workday. The question executives face is no longer whether to adopt it. The question is how to govern it once it sits inside daily work.
One influential answer treats that question as a cultural problem. I want to take that answer seriously, grant most of it, and then show the one question it leaves open. That open question, I will argue, is not about culture, ethics, or oversight. It is about authority: who legitimately decides, and where decision authority should pass between an AI system and a person.
What the Culture Argument Gets Right
In its masterclass write-up AI governance needs more than policies: Why culture will determine success, the ethics and compliance firm LRN makes a clear case. Technical controls alone will not carry an organization through the AI era. The organizations that succeed will be the ones whose culture can govern the technology they deploy.
LRN's argument holds up on six points, and I accept all of them:
- AI governance lags adoption. AI is already embedded in workflows while oversight frameworks remain immature.
- Shadow AI spreads through good intentions. Employees reach for public tools to work faster, and in doing so expose confidential data, intellectual property, and personal information. As LRN puts it, "You cannot govern AI that you cannot see."
- Culture covers what rules cannot. No policy anticipates every situation, so the deciding factor becomes whether employees instinctively ask, "Is this accurate? Is this appropriate?"
- Misplaced confidence beats hallucination as a risk. People stop challenging outputs that look polished.
- Treat an AI agent like a new colleague. Onboard it, expose it to the code of conduct, hold it to standards, rather than testing only for technical accuracy.
- Leaders set the tone. Frame AI as cost-cutting and people resist; frame it as a way to do better work and they engage.
This is the right diagnosis of a real problem. I have little to add to it.
Where Culture Still Leaves a Question Open
Walk through one case. A company adds AI to its mid-career hiring.
The system checks each application for completeness, scores experience against the role, and recommends a shortlist for interview. Culture and AI literacy explain this stage well. A hiring manager who refuses to take the shortlist at face value, who asks whether the screen was fair, is exactly the employee LRN describes.
Then the real question arrives. When the company hires one of these candidates, who decides? Does the manager follow the recommendation, or can she overrule it? If she can overrule it, at what stage, and on whose authority does the reversal stand?
That question is not cultural. It is structural. Underneath it sits a sharper one: where should the line fall between what the AI settles and what a person owns?
A manager with strong judgment still stalls if no one has told her how far her discretion runs. Culture puts people in a state where they can ask the right question. It does not tell them who answers it, or where.
Governance Sets the Frame but Not the Line
Return to AI governance in the formal sense. Governance writes the acceptable-use policy, names the accountable function, and stands up the audit. An organization needs all of it.
Governance operates on the institution. It decides how the organization sets rules and supervises itself. It does not descend to the level of a single decision and mark where one person's authority begins.
So governance can rule that hiring teams may use AI. It does not specify how much of a particular hire the AI may settle and where the manager takes over. Governance is not the weak link here. Its unit of analysis is the institution, not the decision. Judgment Architecture begins where governance leaves authority unspecified.
Human-in-the-Loop Is Necessary but Not Sufficient
The standard response to this gap is Human-in-the-Loop: keep a person somewhere in the flow rather than letting the AI close the loop alone.
Policy is converging on the same idea. Japan's AI Business Guidelines Version 1.2, issued jointly in March 2026 by the Ministry of Internal Affairs and Communications (MIC) and the Ministry of Economy, Trade and Industry (METI), ask organizations that build autonomous AI agents to put mechanisms in place that keep meaningful human judgment in the loop, specifically to contain risks such as unintended actions and privacy violations.
Culture, governance, and Human-in-the-Loop. The toolkit looks complete.
It is not. Inserting a human does not, by itself, define authority, escalation, override, or accountability. A human placed after an irreversible step is a witness, not a decision-maker. The position of the person changes everything: before the agent executes a payment or after, before the shortlist narrows or after. Human-in-the-Loop establishes that a person is present. It does not establish where the person stands, or what they are empowered to do when they get there.
What remains is a design question. Where does decision authority transfer, and to whom?
The Missing Layer
Organizations rarely design authority transfer itself. They leave it to local practice and individual discretion, and they call the result judgment. While AI only assisted at the margins, the ambiguity cost little. Now that AI reaches the edge of the decision itself, the ambiguity is the exposure.
This is the layer that culture, governance, and Human-in-the-Loop all leave untouched. The missing layer is Judgment Architecture, and Decision Design™ is a framework for building it.
Decision Design™ is not about improving decisions alone; it is about designing the authority structure within which decisions become institutionally legitimate. It treats one question as a design object: for a given decision, what does the AI settle, what does a person own, and where does authority pass from one to the other.
The distinction from governance is a distinction of layer, not of quality. Governance manages institutions; Decision Design™ structures the judgment inside them. Governance defines oversight; Decision Design™ defines how authority is allocated. Governance establishes who is accountable; Decision Design™ specifies where accountability transfers between an AI system and a person.
Decision Boundaries™: Where Authority Transfers
The core construct of Decision Design™ is the Decision Boundary™: the institutional point at which legitimate judgment authority transfers between an AI system and a human decision-maker. Decision Boundaries™ are not operational thresholds; they are institutional demarcations of legitimate authority.
A workflow has more than one such boundary. Four types cover most cases:
- Delegation. What the organization hands to the AI. Document checks and first-pass analysis, where the work is routine and a person can verify it.
- Escalation. The conditions that return a decision to a person. Set them in advance: the amount clears a threshold, the case has no precedent, or two stakeholders disagree.
- Override. Whether a person can reverse the AI's output, and if so, who holds that power and under what authority.
- Suspension. Whether anyone can halt the AI when it misbehaves, who holds the stop, and who owns the work once it stops.
Naming these four lines is what designing a Decision Boundary™ means. Leave them implicit and the organization has not designed its boundaries; it has merely deployed a system and hoped.
A Worked Example: AI-Assisted Hiring
Take the hiring case again and draw the boundaries explicitly.
- Intake. The AI receives applications.
- Screening. The AI checks each application for completeness. Delegation: routine and verifiable, so the organization assigns it to the AI.
- Fit analysis. The AI scores experience against the role and proposes a shortlist. Delegation, but a proposal, not a decision.
- The Decision Boundary™. Draw the line here. If scores cluster too tightly to separate, a résumé carries a gap that needs explanation, or the hiring manager and the line owner disagree, the case escalates to a person.
- Human decision. The hiring manager and the line owner read the recommendation and its basis, then decide. They can override the AI.
- The Decision Log. Record who decided, on what grounds, and why. Where the human decision departs from the recommendation, record the reason.
- Notification. Confirm the outcome and notify the candidate.
Nowhere in this flow can anyone say "the AI decided." The AI reaches the edge of the decision; the person past the Decision Boundary™ owns it.
The Decision Log is not minutes of a meeting. Decision Logs do not merely record outputs; they preserve accountability continuity across distributed judgment processes. It records not what the AI recommended but who took responsibility for the judgment, which is what lets an auditor later trace whose decision it was and confirm the boundary held.
How Decision Design™ Sits Alongside Existing Disciplines
Decision Design™ is a framework for Judgment Architecture, and it does not replace the disciplines around it. Each solves a real and separate problem, and Decision Design™ works across them.
- Governance sets the institutional frame. It does not mark where one decision passes from AI to person.
- Digital transformation redesigns the work. It does not say who decides inside the new process, and efficient processes often hide the decision rather than clarify it.
- Automation speeds the task. It does not assign responsibility for the outcome when the task goes wrong.
- AI ethics defines what counts as fair, transparent, and accountable. It does not specify who applies that standard, or where.
Each supplies something Decision Design™ needs. None of them draws the Decision Boundary™. That line is the work Decision Design™ does.
Beyond Better Judgment
This article opened with the claim that AI governance will be decided by culture. Half of that is right. Culture builds people who ask the right question and refuse a convincing but wrong answer. It does not decide who owns the question, or where authority for it sits.
The AI era does not merely require better judgment from individuals. It requires organizations to design the structure within which judgment becomes legitimate: who decides, who can override, who escalates, who can stop the system, and where accountability transfers. The guidance now emerging from regulators asks for human judgment. Human judgment becomes operational only once an organization has designed where it lives.
AI will continue to participate in judgment. Whether organizations remain accountable will depend on whether they deliberately design the authority structures within which judgment becomes legitimate.
Decision Design is a judgment architecture framework proposed by Ryoji Morii, founder of Insynergy Inc., for structuring authority, accountability, and decision boundaries in AI-augmented organizations.
References (Primary Sources)
- LRN. AI governance needs more than policies: Why culture will determine success. Key insights from the ICA & LRN Masterclass: Managing Risk, Culture and Behaviour in the Age of AI.
- Ministry of Internal Affairs and Communications (MIC) and Ministry of Economy, Trade and Industry (METI). AI Guidelines for Business, Version 1.2. March 2026.