Over the past two decades we have digitalised processes, standardised touchpoints, and tamed complexity with workflows and rules. And yet the services that truly matter - those full of exceptions - continue to elude rigid systems. Enter AI agents, shifting the focus from executing predefined steps to achieving a user-declared goal: enrolling a daughter in school, obtaining a reimbursement, organising a trip with multiple constraints. The article by Nick Babich captures this paradigm shift well, highlighting the key traits of the new “agentic” service design scenario. UX Planet
Why now
This isn’t just hype: market forecasts point to a fast “agentification” of enterprise software and customer service, albeit with setbacks and reassessments along the way. For instance, Gartner estimates that by 2026 a significant share of enterprise apps will embed task-specific agents; at the same time it warns that over 40% of agent-based projects may be cancelled by 2027 due to cost and limited value - a clear sign of the risk of “agent washing”. In short: big opportunities, but they require discipline in design, governance and impact measurement.
In parallel, Generative UI (interfaces created in real‐time by AI, tailored to intent, context and user competence) is pushing toward an outcome-oriented design: fewer fixed screens, more “scenes” constructed when needed. The Nielsen Norman Group describes this as a shift in priority: design goals and constraints so the AI composes the right interface at the right time.
What really changes in Service Design (from practice)
Speaking from my experience as a UX & Service Designer working through full cycles of discovery, blueprints and pilots:
From journey to mission. It’s no longer “guide the user through 12 screens”, but rather articulate the user’s intent and constraints, then let the agent orchestrate across channels and systems. User stories turn into negotiated outcomes with acceptance thresholds (e.g., cost, time, reliability). Babich identifies this goal-oriented autonomy as a pillar of the new service design. UX Planet
From human-centric UI to agent-centric APIs. If the agent must act on the user’s behalf, organisations must expose reliable services via APIs/authorisations with clear contracts and logging. Companies that become “AI-friendly” in these integrations will win in speed and perceived quality. Enterprise adoption forecasts confirm the trend.
From static UI to G-GUI (Generative GUI). No more endless wizards for every exception: the interface is built on-demand, centred on intent, state and trust (what is the agent doing? with what sources? within what limits?). The NN/g concept of Generative UI enables dynamic and personalised experiences - provided guardrails are well defined.
From technical exclusivity to democratization. Tools like builders and no/low-code environments accelerate the creation of vertical agents. The front-end and ideation itself are “assisted”: for example, tools that generate UI layout and code from prompts (e.g., Google “Stitch”). Great for rapid prototyping, but evaluative quality must remain high.
From compliance to “designed trust”. Agents touch money, data, reputation. Guardrails aren’t optional: action-limits, spending thresholds, mandatory human checks on irreversible acts, verifiable reasoning, audit trails and dry-run modes. Analysts expect benefits in customer service - but only with serious governance.
A concrete case
What I ask: “Enrol my daughter in the local school by Friday. Max commute time 15 minutes. If no place, propose nearby alternatives.”
What the agent does:
Accesses portals via SPID/CIE and municipal/ministry systems.
Verifies residence, family status, siblings already enrolled.
Checks available schools and actual travel time.
Populates the application with known data; asks only for missing items (e.g., ISEE, certificates).
Activates lunch service/transport and handles payment via pagoPA.
Logs everything and alerts only when my confirmation is needed.
What I see on screen:
A simple dashboard showing:
Application status (draft → submitted → registered)
Next steps by the agent
Confirmation buttons (privacy, submit, payment)
Alternatives if class is full, listing pros/cons (distance, services)
Practical toolkit
Outcome Map > User Flow. Map intents, constraints, acceptance thresholds, preferred fallback.
Agent-Blueprint. A version of the service blueprint with lanes for API/identity/consent, decision points, spending policy and confirmation gates.
Contract-first API. Versioned, testable schemas with rate limits and error policies; define minimal capabilities so agents can operate securely.
Explainability by design. Every agent action has a brief “why”, source, log and status; dry-run mode available to preview actions before execution.
Generative UI sensibly. Interfaces “just-in-time” but consistent in key patterns (terminology, feedback types, critical commands). NN/g: outcome-oriented, not pixel-driven.
Double-bottom metrics. Measure user outcome (time, stress, resolution) + business outcome (costs, NPS, compliance). Efficiency projections exist, but must be validated in context; don’t take analyst numbers as gospel.
Where to start tomorrow (checklist)
Choose a process rich in exceptions (not the easiest one).
Write the Intent Spec: objectives, constraints, acceptance thresholds.
Design three non-negotiable trust gates.
List five critical APIs and the data/consent inventory.
Prototype a supervision view (status, next moves, “why”).
Measure a baseline and set up an A/B test with agent-assisted flow.
Prepare a fallback plan (human switch-back).
Document failure modes and fallback behaviours.
Conclusion
AI agents force us to change our posture: from guiding every click to negotiating outcomes and trust. For designers, this means shifting focus to intents, constraints, APIs and governance, with interfaces that emerge when needed and explain what the agent is doing for us. The future of services - especially those “too complex to be designed as screens” - will be played here.
References & Further Reading
Nick Babich, Service Design in the Era of AI Agents (UX Planet, Oct 9 2025). UX Planet
Gartner, predictions on “agentic AI” (Mar-Aug 2025) and project cancellation risk by 2027.
Nielsen Norman Group, Generative UI and Outcome-Oriented Design (March 2024) and ongoing resources on AI in UX.
Google “Stitch”: UI code generation from prompts (May 2025).
