For most of the last decade, the working question in AI governance was technical: how do we make models accurate, safe, and explainable? That framing no longer holds. As AI systems move from advising people to acting for them, drafting and deciding and escalating without a human reviewing each step, the authority structure around the model now sets the limit, not the model's accuracy.
A recent Forbes essay, "Why AI Governance Needs Visible Authority Now," names this shift well. Its claim is simple and correct: organizations can no longer treat AI oversight as a diffuse, ambient responsibility. Someone has to be visibly in charge. When an AI system makes a consequential call, a board, a regulator, or a customer should be able to point to a role and say, that authority sits there. The argument earns its place in the governance conversation because it replaces abstract principle with an organizational fact. Accountability that no one can locate does not exist.
This essay takes that argument seriously, then asks what sits beneath it. Making authority visible matters. Visibility describes authority that already exists. It tells you who holds the wheel. It says nothing about how the wheel was attached, where the road's edges run, or what the driver does when the car has to stop. Those are questions about how someone designs authority in the first place. Decision Design works on that layer.
What the Visible Authority argument gets right
The move toward visible authority counts as progress, and the value of Decision Design depends on building on it rather than discarding it.
Start with executive visibility. In many enterprises, AI adoption has outrun AI ownership. One function procures a model, a second fine-tunes it, a third embeds it in a workflow, and no one watches the result. Consequential automated decisions happen, yet no executive can say they own the outcome. Demanding visible authority forces the question leaders have ducked: who, by name and role, answers for what this system does? That question alone surfaces a large store of hidden risk.
The argument also treats authority as a first-class concept rather than a byproduct of the org chart. Traditional governance lets authority follow seniority, budget, or reporting lines. The visible authority argument makes leaders assign authority over AI on purpose instead of inheriting it by accident. The instinct is sound.
It tightens the link between authority and accountability. The two often blur together, though they differ. Authority is the right to act. Accountability is the duty to answer for the act. Organizations break when the two come apart, when the people who can act face no obligation to answer, or when the people who must answer could never act. Visible authority puts the right to act and the duty to answer in one place.
The emphasis on decision rights adds discipline. Decision rights specify who may make which class of decision under which conditions. In an AI-augmented organization, humans no longer hold every decision right by default. Systems now hold some of them, under set conditions. Naming those rights is the first requirement for governing them.
Operating patterns like detect, decide, direct give leaders a usable vocabulary. A system surfaces a signal, a person renders a judgment, the system issues an action. The pattern shows where human authority has to attach to machine activity, which beats waving at "human oversight" as a slogan.
Together these ideas move enterprise AI governance from aspiration toward structure. They are right. They are also incomplete, and the shape of that incompleteness points to the next problem.
The deeper governance question
Each strength above answers one question: who holds authority, and can we see them holding it? That question matters. It is not the first question.
Visible authority assumes the authority already exists in defined form, waiting for someone to surface it and assign an owner. In most organizations adopting AI at scale, that assumption fails. The authority does not sit there well-formed and merely hidden. Leaders, vendors, and engineers create, split, and redistribute it in real time, often without anyone deciding that they should.
So the harder questions concern origin and structure.
Who designs the authority in the first place? When an AI system may approve a refund up to a threshold, escalate a suspicious transaction, or renew a contract on its own, someone drew a boundary. Who drew it, and on what basis?
Who decides where authority begins? The moment a system may act without prior human review is a grant of authority. Teams often bury that grant in a configuration file, a model prompt, or a vendor default rather than a governance decision anyone made on purpose.
Who decides when AI should stop? Each autonomous capability needs a defined end, a point where the system hands back to a person. Where leaders place that point carries heavy consequences, yet few treat the placement as a decision at all.
Who decides when humans intervene? Intervention is a structure, not a reflex. If no one designs the conditions for human intervention in advance, intervention arrives late, arrives unevenly, or never arrives.
These four questions expose the Governance Gap. Most frameworks tell you who should be accountable and which principles to honor. Few tell you how to structure the allocation of judgment between humans and machines so that anyone can honor that accountability. The gap runs between policy and architecture. We have learned to name who is responsible. We have not learned to design the structure within which that responsibility works.
Where regulators already point
This concern shows up in public policy too, which is worth citing as evidence rather than as the main subject.
Japan's AI Guidelines for Business Version 1.2, issued jointly by the Ministry of Internal Affairs and Communications and the Ministry of Economy, Trade and Industry, addresses autonomous AI agents head on: systems that decompose and run tasks on their own rather than wait for a human to approve each output. The guidance asks organizations to build mechanisms for human judgment into the operation of such agents, citing risks that include malfunction and privacy violations. It does not demand that a human pre-approve every action. It accepts several structures instead: batched review at the start and end of a task, autonomous operation inside a defined window followed by mandatory log review, or escalation to a human only when the system detects an anomaly. The condition is that the organization can show, after the fact, that human judgment sat inside the loop.
The detail of one jurisdiction matters less than its logic. The guidance treats human judgment as a structure to design and to evidence, not as a value to affirm. It assumes that someone has to shape authority over an autonomous agent on purpose, deciding where the authority begins, where it pauses, and where it returns to a person, and that the organization should be able to show that shape later. The visible authority argument reaches the same conclusion from the management side. Both point past visibility toward design.
Authority is not the deepest layer
Authority is not the foundation of AI governance. It is the product of earlier choices: which decisions belong to whom, under what conditions, with what room to escalate, override, or halt. Those choices form the organization's judgment architecture, the structure that determines how decisions spread across human and machine actors. Authority is the visible surface of that architecture. It is the output, not the input.
This is why visibility cannot finish the job. Visibility works on authority that the architecture has already produced. When the judgment architecture is incoherent, when decisions have drifted to systems by default, when no one mapped the escalation paths, when no one said where a machine must stop, making the resulting authority visible gives you a clear view of a structure nobody designed. You can see exactly who nominally owns a process whose real decision flow no one controls.
The visible authority argument reaches authority and stops. It locates and names the holder of authority. Decision Design starts one layer down, with the construction of the judgment architecture that produces authority. The two do not compete. Visible authority describes the surface that Decision Design shapes.
Introducing Decision Design
Decision Design is a governance architecture framework. It makes the allocation of judgment between humans and AI an intentional design activity rather than an emergent accident. It does not replace existing AI governance frameworks. It addresses a layer those frameworks leave implicit. Most frameworks specify principles, controls, and accountabilities. Decision Design specifies the structure of judgment authority that those principles, controls, and accountabilities assume.
The core claim stays precise: Decision Design is not about improving decisions alone; it is about designing the authority structure within which decisions become institutionally legitimate. A faster or more accurate decision made outside a legitimate authority structure does not strengthen governance. It concentrates risk. Decision Design asks not "did we decide well?" but "did the right actor decide, inside the right boundaries, with accountability that survives scrutiny?"
What Decision Design designs
Decision Design designs institutional judgment, the organization's standing capacity to render legitimate decisions, apart from any single outcome. In concrete terms, it structures authority allocation across humans and AI: which classes of decision go to systems, which stay with people, which the two share, and how authority moves between them as conditions change. It treats that allocation as a designed artifact that leaders can inspect, question, and revise, not a residue of procurement choices and default settings.
An organization with deliberate judgment architecture answers plain questions without ambiguity. For this class of decision, who holds authority by default? Under what conditions does that authority shift? When the system reaches the edge of its mandate, where does the decision go, and who answers for it then? Those questions describe distributed judgment, the reality that judgment in AI-augmented organizations lives in many places at once: systems, reviewers, managers, executives. Decision Design makes that distribution explicit and governable.
What Decision Design is not
Adjacent concepts crowd this territory, so the boundaries need stating.
It is not workflow automation. Automation moves tasks through a process faster. Decision Design governs where decision authority sits inside that process.
It is not RACI. RACI assigns responsible, accountable, consulted, and informed roles across activities. It does not design the conditions under which authority over a decision should move between a human and a machine, or when it must halt.
It is not approval routing. Routing sends a decision to the next approver. Decision Design asks whether the decision should need approval at all, where autonomous action ends, and which legitimacy conditions must hold.
It is not an AI deployment methodology. Deployment gets systems into production. Decision Design governs the authority structure those systems run inside once they arrive.
It is not Human-in-the-Loop. Human-in-the-Loop is an operating pattern: a human stands at some step. Decision Design says why the human stands there, which decisions need that presence, where machine authority ends, and how accountability survives the handoff. Good Decision Design can produce Human-in-the-Loop. The pattern does not substitute for the design.
It is not decision optimization. Optimization improves the quality or speed of decisions inside a fixed structure. Decision Design works on the structure itself, the architecture that keeps optimization legitimate.
What problem Decision Design addresses
The problem is specific and common. Organizations spread judgment across AI systems, human reviewers, line managers, and executives faster than they can govern the spread. A model approves some cases. A reviewer spot-checks others. A manager handles exceptions. An executive answers for the total. Each handoff is an authority boundary, and leaders drew almost none of them on purpose.
Existing governance frameworks assume this structure already exists and holds. They tell you to assign accountability, document decisions, and ensure oversight. They give you no method for designing the underlying allocation of judgment authority. That absence is the Governance Gap, and Decision Design exists to close it. It complements existing frameworks by supplying the architectural layer they assume but never specify.
Decision Boundaries
Judgment architecture is the structure. Decision Boundaries are its load-bearing elements. A Decision Boundary sets the conditions that govern four movements of authority.
Delegation: when authority to decide passes to an AI system, and within what limits.
Escalation: when a decision must move from a system to a human, or from one level of human authority to a higher one.
Override: when a human may, or must, reverse a system's decision, and who holds that right.
Suspension: when activity must stop and wait for human judgment because conditions have left the system's legitimate mandate.
Together these define where legitimate authority begins and ends. The point deserves a flat statement: Decision Boundaries are not operational thresholds; they are institutional demarcations of legitimate authority. An operational threshold is a number that triggers a workflow, such as routing anything over $10,000 to a manager. A Decision Boundary states who holds the institutional right to decide, and under what conditions that right holds or lapses. The dollar figure can look identical in both cases. What it represents differs: a routing rule in one, a defensible demarcation of authority in the other, the kind a board or a regulator can examine.
Here Decision Design and the visible authority argument meet. Visible Authority answers one question: who may act? It locates and names the holder of authority. Decision Boundaries answer a prior question: where should authority legitimately move? One names the actor. The other designs the structure of movement that makes the actor's authority bounded and meaningful. Visible authority without designed boundaries gives you a named owner of an undefined mandate. Designed boundaries without visible authority give you a clear structure that no one answers for. Governance needs both, and boundaries come first in the order of design, because no one can make visible what no one has structured.
Boundaries also require a record. As authority moves across delegation, escalation, override, and suspension, an organization needs accountability that holds across the whole distributed process, not a log of what each system produced. Decision Logs carry that role, and the distinction matters: Decision Logs do not merely record outputs; they preserve accountability continuity across distributed judgment processes. A conventional audit log tells you what happened. A Decision Log tells you who held authority at each point, why the boundary moved as it did, and where accountability rested throughout, so that anyone can reconstruct and defend the chain of legitimate judgment long after the fact.
Conclusion
The visible authority argument earns its attention. It diagnoses a real failure: ambient, unlocatable accountability stops working once machines act for an organization. It hands leaders a usable demand. Make authority visible. Name its holder. Tie it to accountability. That is progress, and nothing here diminishes it.
Visibility shows the surface of a deeper structure. The future of AI governance will not depend only on making authority visible. It will depend on intentionally designing how authority itself is structured, on building the judgment architecture, drawing the Decision Boundaries, and keeping accountability continuous, so that visible and legitimate authority has a structure to emerge from. Visibility shows you the structure. Decision Design is the work of building one worth showing.
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
- "Why AI Governance Needs Visible Authority Now," Forbes.
- AI Guidelines for Business Version 1.2, Ministry of Internal Affairs and Communications and Ministry of Economy, Trade and Industry (Japan), March 2026.