Should designing an AI shopping agent be its own job, or just part of designing the whole customer experience?
Why should service design absorb agent design rather than keep them separate?
This explores why some designers argue that designing AI agents shouldn't be its own specialty, separate from designing the whole service a customer moves through, and what the corpus offers for and around that claim.
This explores why designing an AI agent might belong inside service design (the work of shaping a whole customer journey, its back-office processes and its handoffs) rather than sitting alongside it as a separate discipline. The most direct argument comes from Dubberly. He treats LLMs as a new design material and expects designing for agents to absorb interaction design and become central to service design, in the same way earlier technology shifts reshaped design disciplines Is design for agents becoming the next major design paradigm?. Only that one note makes the claim outright. The other material supports it from the side: it shows what goes wrong when agents are designed in isolation from the system around them.
The practical case is that an agent added on top of an existing service tends to fail. Deloitte finds that many agentic AI projects stall because organizations layer agents onto old workflows instead of redesigning the end-to-end process, which means rethinking data, orchestration and management together Does agentic AI succeed through process redesign or workflow layering?. Process redesign of that kind is exactly what service designers do. Fin's team adds a concrete piece of evidence: when someone deliberately owned the agent's tone, level of detail and handoff to humans, customer satisfaction rose measurably. One warmer opening message lifted it from 72.8% to 78.4% Does conversation design ownership measurably improve AI agent performance?. The handoff is the telling part, because it is where the agent stops and the rest of the service begins. You can't design that moment well from only one side of it.
There is also an older lesson that predates LLMs. Multiagent-systems research from the early 2000s kept finding that you can't design agents well by designing only the agents. Gaia argued that the environment the agents act in has to be modeled explicitly, or the system as designed drifts away from the system as it actually runs Should multiagent systems model their environment explicitly?. Gaia also argued that organizational rules, such as who may join and which constraints span several agents, can't be derived from individual roles Should organizational rules be designed separately from agent roles?. MOISE+ went further and split the design into structure (roles and groups), function (shared plans) and permissions and obligations Can separating structure and function improve multiagent organization design?. Read today, these look like service blueprints: they describe the organization the agent works within, not just the agent.
The same pattern shows up in current agent engineering. Walden Yan argues that parallel subagents break down because each one makes unwritten assumptions the others can't see Why do parallel subagents produce conflicting design decisions?. That is a small-scale version of separately designed parts that don't fit together. Dubberly's own layer model places agents inside supervisor and orchestrator hierarchies, so oversight becomes a design layer in its own right How should agent capabilities be organized into functional layers?. On the alignment side, researchers argue that the agent-serves-one-user framing is too narrow and that agents have to balance several stakeholders, with accountability and transparency built in Should AI agents serve one user or balance many interests?. Balancing several stakeholders is a service-design problem, not something you can settle inside the agent.
The idea you might not expect is that this absorption may flow in both directions. UI-Verse found that interface changes aimed at AI agents, such as clearer labels and more predictable controls, helped agents complete tasks without making the interface worse for humans Can interface design help agents complete computer tasks better?. If one design can serve both kinds of user, that strengthens Dubberly's case. The service becomes the thing being designed, and agents become one more kind of participant in it, sometimes the company's agent and sometimes the customer's.
Sources 10 notes
Dubberly argues design is entering a new paradigm driven by LLMs as a design material, where agent design will subsume interaction design and become core to service design. He grounds this in design history's pattern of technological shifts reshaping design disciplines.
Deloitte argues many agentic AI implementations fail because organizations layer agents onto old workflows rather than redesigning operations. Success requires agent-compatible data architectures, robust orchestration, and new management models; HPE and Toyota illustrate this approach.
Fin's Blueprint Team found that intentional conversation design—controlling tone, detail level, and handoff structure—measurably improves outcomes. A single warm opening message raised CSAT from 72.8% to 78.4% in A/B testing.
Gaia's 2003 methodology argues that MAS designers must model environments (physical or virtual) as primary abstractions during analysis and design. Skipping this step causes complexity and misalignment between intended design and actual operation, as illustrated by a conference-management system example.
Gaia's 2003 methodology shows that role and interaction models alone cannot capture constraints that span multiple agents or governance decisions about system membership. Organizational rules and structures must be first-class design abstractions, not implications of role models.
Show all 10 sources
Hübner, Sichman, and Boissier's 2002 model splits organizational specification into structural (roles and groups), functional (global plans), and deontic (permissions and obligations) dimensions. This separation avoids failures where agents lack guidance or lose validated plans.
Walden Yan argues multi-agent parallelism is fundamentally fragile: subagents embed unwritten assumptions in every action, so agents working independently converge on conflicting choices. The fix is reverting to single-threaded linear agents where context remains continuous.
Dubberly proposes a two-axis stack model: a capability axis (chatbot → system prompt → RAG) adding inputs that extend what LLMs can act on, and a scale axis (monitoring agents → supervisors → orchestrator) aggregating oversight. He cautions that ubiquitous agent hierarchies raise design risks alongside capability gains.
Research argues the traditional single-agent-single-user alignment framing is insufficient at scale. Instead, governance should rest on inclusivity (balancing diverse interests), visibility (rigorous tracking and transparency), and liability (developer and deployer accountability for harms).
UI-Verse benchmark testing found that augmented heuristics targeting agent perception, grounding, and control consistently improved agent task completion and efficiency. Human studies confirmed these designs preserved usability for human users with no observable regressions.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Drop the Hierarchy and Roles: How Self-Organizing LLM Agents Outperform Designed Structures
- A Model for the Structural, Functional, and Deontic Specification of Organizations in Multiagent Systems (MOISE+)
- Developing Multiagent Systems: The Gaia Methodology
- Embodied LLM Agents Learn to Cooperate in Organized Teams
- Toward an Organizational Science of Multi-Agent LLM Systems: Decoupling Who, How, and Which Algorithm
- Comparing Human Oversight Strategies for Computer-Use Agents
- Explaining AI Agents Through Execution Traces
- Worse Together: How Performance Breaks Down in Multi-User Multi-Agent Teams