Six months in, the program review. Fourteen agents in production. Each one demos beautifully. None of them know about each other.
The claims-intake agent doesn't know the underwriting agent exists. The compliance team has built its own. The contact-center team has built its own. Two of them write to the same case-management system; nobody can say which one wrote what.
The CFO asks the question every CFO asks at month six: where is the P&L impact? The answer is in the room, and nobody can say it.
You built fourteen agents and zero services.
This scene is repeating, with minor variations, inside every regulated enterprise that began an AI program in the last twenty-four months. The agents work. The transformation does not. The reason is not the agents.
The reason is that nothing was designed around them.
What's missing has a name. It has six names, actually — six disciplines that, taken together, design the operational service the AI is supposed to enable. Each one answers a fear the buyer has already said out loud in a quiet room. Each one was missing from the program review above. We are publishing this magazine to argue that experience design — the discipline UX Magazine has covered for twenty years — is the only practice that contains all six in working order, and that without all six, agentic AI does not become transformation. It becomes inventory.
You built fourteen agents and zero services. The difference between those two sentences is the entire field.
Service design
"Why our AI investment keeps producing demos instead of P&L."
The first thing missing from the program review was a service. Each agent solved a slice of a workflow. None of them was a redesign of the work itself. Service design is the discipline that asks what the operational service actually is — what the citizen receives, what the claimant receives, what the patient receives — and then designs that service end to end, with the agents inside it.
This is the practice that makes "transformation" mean something specific to a director or a VP. It is not a slide. It is a redesign of how claims intake or scheduling or benefits enrollment works as a service, with the human moments and the machine moments arranged so the service is faster, cheaper, and better than the one it replaces. Without service design, an agent program produces demos because each agent is the answer to a question nobody asked at the service level. With service design, the agents are subordinated to the service, which is what the P&L was always measuring.
Information architecture
"Won't incremental deployment just produce a mess?"
The second thing missing was a shared model of what each agent knew, when, and from where. Information architecture is the discipline that structures knowledge, context, and operational state so that the agents compose instead of sprawl. It is the architectural answer to the diligence question every CIO asks: if we deploy one agent at a time, won't we end up with fourteen agents and six context models?
Yes. You will, unless somebody designs the information substrate that the agents stand on. Information architecture decides what a "case" is, what an "account" is, what "consent" is, and what facts are authoritative under what conditions. When this discipline is present, an incremental rollout compounds — each new agent makes the next one easier to build, because they all read from and write to the same coherent model of the work. When this discipline is absent, every new agent is a new island. The mess is not produced by incremental deployment; it is produced by incremental deployment without architecture.
Interaction design
"Why approvers won't actually use the approval queue we built."
The third thing missing was a usable handoff. Every regulated-enterprise agent program eventually produces a human-in-the-loop queue: a list of agent-generated proposals waiting on a person to approve, reject, or modify. In most programs, that queue is unworkable. It has too many items, too little context, too little trust, and approvers route around it within weeks.
Interaction design is the discipline that makes the handoff between human and agent something a person will actually do. It decides how a proposal is summarized, what evidence accompanies it, what the approver is being asked to attest to, and how trust is earned over time as the system demonstrates restraint and accuracy. Without this discipline, "human-in-the-loop" is a compliance gesture that nobody runs. With it, the loop closes — and the agent's authority grows in the only way agent authority can legitimately grow in a regulated environment, which is by an approver expanding it.
Conversation design
"Why employees and customers route around the agent we launched."
The fourth thing missing was a dialogue that worked. Conversation design has lived inside contact centers for a decade, optimizing scripts and intent matchers. The agentic era moves it to the center of the enterprise, and the stakes change. A chatbot that frustrates a customer is a brand problem. An agent that frustrates the underwriter, the case manager, or the compliance officer is an operational problem — because they will go around it, and the program metrics will quietly collapse.
Conversation design in this register is not about clever prompts. It is the discipline that decides how dialogue is structured across voice, chat, and multimodal channels so that the person on the other end stays in the conversation. It governs turn-taking, repair, escalation, voice persona, and the precise moments when the agent should stop talking and act, or stop acting and ask. The diagnostic question is simple and brutal: are your people going around this thing? If yes, it does not matter what the dashboard says.
Systems and operational design
"Can we start with one workflow and grow without re-platforming?"
The fifth thing missing was a runtime that composed. Systems and operational design is the discipline that decides how the orchestration, the composition primitives, and the agent runtime are structured so that the platform itself is coherent. It is the discipline that answers the buyer's most defensive question: if we start small, can we grow without ripping everything out?
The answer is "yes, if the platform was designed to be incremental and comprehensive at the same time." That is not a marketing claim. It is a property of the runtime — whether agents can be authored, deployed, governed, and combined without ad-hoc glue every time the program scope expands. When this discipline is missing, the second workflow re-platforms the first. When it is present, the eighth workflow takes a week, not a quarter. This is the operational difference between a platform program and a string of pilots.
Organizational design
"What happens to our org chart when agents take 30% of the workflow?"
The sixth thing missing was a theory of the people side. Most agent programs treat the org chart as a fixed input — the workflow gets redesigned, the agents get deployed, and somebody figures out the human consequences afterward. Then the program meets the union, the line manager, the controller, the auditor, and the head of risk, and the program slows or stops.
Organizational design is the discipline that decides how the customer's teams, roles, and workflows evolve as agents take on work. It names who manages an agent. Who fires one. Who owns the decisions an agent encodes. Who is accountable when it fails. Which roles disappear, which roles change, which roles emerge. This is the piece that every platform vendor — including the ones with dramatic government contracts — leaves out of the engagement, and it is the piece without which the other five disciplines arrive at a workforce that has no theory of what to do with them. Real transformation is human transformation. The agents are the easy part.
An agent is technology. A service is a promise. An organization is the place where the promise is kept.
What this magazine is for
UX Magazine has been publishing for twenty years. For most of that time, the field expanded outward — from the screen, to the customer experience, to the employee experience, to the operating system of the institution itself. The agentic era is the next move outward, and it is the most consequential one, because for the first time the system on the other side of the design is capable of acting.
Six disciplines make that work. The reason to name them together — and to make them the structural taxonomy of this publication going forward — is that they have to be present together, and almost everywhere in industry today they are not. Each section of UX Magazine is now organized around one of them. Each article we publish lives in the column for the discipline whose absence would have produced the problem the article describes. The argument is not in a manifesto on a separate page. The argument is the publication.
This is a living taxonomy. Each issue of UX Magazine fills the columns forward. We are not going back to retag the past — the past is what made the present columns visible.