Best Practice

Designing Agentic Experiences

Spark Reply's approach helps complex organisations in architecting intent-driven human-agent collaboration across the employee and customer experience.

Boosting the Adoption of Agentic Systems

Generative AI operates by generating content in response to prompts, typically managing one interaction at a time. Agentic AI operates differently by working towards a defined goal, planning a sequence of actions, and connecting to external systems such as customer relationship management (CRM) platforms, enterprise resource planning (ERP) systems, and databases.

It reflects on outcomes and executes actions continuously without requiring human input at every step. Agentic artificial intelligence pursues goals, plans multi-step actions, and orchestrates entire business processes. Designing these experiences effectively is the differentiating factor between systems that empower users and systems that overwhelm them.

The Enterprise AI Market is Evolving Quickly

Organisations are rapidly advancing through three maturity stages of artificial intelligence adoption.

Generative
AI assistants

Reactive tools functioning as embedded helpers that generate content in response to a user prompt within existing workflows.

AI agents as
task executors

Proactive agents capable of executing defined tasks within a specific process step without continuous human input, operating within explicit organisational boundaries.

AI agents orchestrating
enterprise processes

Networks of proactive agents that collaborate across multi-step workflows, invoke one another, update shared external systems, and coordinate complex actions.

The Central Design Challenge:
Complexity Must Not Reach the User

As agentic systems become more autonomous, the underlying complexity increases significantly. This encompasses technological complexity in integrating tools, interaction complexity in managing non-linear user journeys, and governance complexity regarding permissions, auditability, and regulatory compliance.

A critical component often overlooked is the orchestration layer. The visible agent on a user interface is frequently a front-facing orchestrator that coordinates a network of specialised back-end agents. These access data and trigger actions in connected systems before presenting results to the user. Treating the front end in isolation typically yields results comparable to a classic chatbot. The underlying complexity must be intentionally designed out of the user's path; otherwise, it shifts directly to the user.

The Foundations of Intentional Design

To mitigate complexity, four foundational parameters must be defined prior to development.

Experience Initiation

Human-Agent Collaboration

Control Boundaries

Recovery Mechanisms

Four Characteristics of a Well-Designed Agentic Experience (AX)

To successfully abstract underlying system complexity and ensure seamless human-agent collaboration, a well-designed Agentic Experience possesses specific defining characteristics.

Multimedia-native

It understands and generates content across formats, combining text, voice, images and video. This enables more natural interactions, improves accessibility and strengthens contextual understanding.

Intent-driven

It separates the experience layer from rigid process flows. Orchestration is based on user intent, not predefined step sequences, which limits the impact of process changes on the user experience.

Natively multimodal

It supports text, voice and graphical interfaces within a single unified interaction model. There is no need to build separate interfaces per modality. The same interaction logic applies across them, ensuring continuity between the front end and the back end.

Natively omnichannel

It integrates seamlessly across channels and touchpoints. Channels act as containers for the experience rather than separate logic layers, reducing development effort and making expansion to new channels faster.

When Do Agentic Experiences Deliver Real Value?

Employee Experience

Increasing Individual Effectiveness
Agentic tools assist individual employees in executing tasks faster and making informed decisions. Applications include personalised daily briefings, smart search functionalities, meeting copilots, email triage, and proactive deadline management.
Fostering Cross-Functional Collaboration
Agentic systems coordinate shared workflows between distinct departments, managing handoffs and maintaining alignment. Applications include human resources case management, ticket triage, and shared workspaces featuring live task visibility.

Customer Experience

Increasing Conversion and Transaction Success
Agents support commercial interactions ranging from assisted selling and product explanations to configuration and quoting. The experience integrates pre-sales and sale workflows while providing embedded customer support with human escalation capabilities.
Improving Retention and Lifetime Value
Agents manage the complete customer lifecycle, including onboarding, ongoing service resolution, and renewals. They deliver proactive nudges and personalised conversational guidance to reduce friction across touchpoints.

Reply’s Methodology for Designing Agentic Experiences

Spark Reply leverages deep expertise in user experience design combined with hands-on knowledge of agentic system architecture and prompt engineering to deliver full-stack orchestration for both single-agent and multi-agent enterprise systems. Spark Reply’s methodology for Agentic Experiences design adapts established design practices to the specific demands of agentic systems, spanning autonomy, orchestration logic, and omnimodal interaction. The approach encompasses four distinct phases.

Framing and Scoping

This phase defines the targeted business objectives, maps user intents by priority, establishes explicit autonomy boundaries, and plans system integrations. The primary output is an AX Strategy Blueprint, which documents the collaboration model and represents agents as named actors with defined roles, decision rights, and escalation paths.

Design and
Implementation

This phase translates the strategic blueprint into tangible human-agent interaction patterns, defining workflow interactions and autonomous action thresholds. It encompasses two tracks: Agent Foundations and Agent Experience. Behavioural parameters such as reasoning style, tone of voice, uncertainty handling, and proactivity levels are defined and encoded into the system prompt, which serves as a technical specification.

Testing and
Refining

The system prompt is not a final design artefact, but a dynamic component that requires continuous testing and iterative refinement to ensure the agent accurately addresses the full range of user intents, behaves appropriately at the boundaries of its scope and handles unexpected inputs without degrading the experience. Spark Reply works with a purpose-build platform for agent testing and performance monitoring to support this phase.

Deployment and
Continuous Improvement

Deployment initiates an operational lifecycle requiring continuous performance monitoring. Standard user experience metrics are insufficient for autonomous systems. Consequently, Spark Reply defines AX-specific key performance indicators (KPIs) prior to deployment, covering metrics such as intent resolution accuracy, autonomous action rates, escalation frequency, and user-perceived trust. This phase also includes adoption strategy and change management.

Frequently Asked Questions

What makes agentic experience design different from conventional UX design?

Conventional UX design works within defined process flows where screens lead to screens and steps follow steps. Agentic experience design operates in an intent-driven model where the experience layer is decoupled from the underlying process. This creates new interaction patterns that have no precedent in traditional product design, such as proactive agent interventions at contextually appropriate moments, uncertainty handling when data is incomplete, negotiation of autonomy boundaries, real-time anomaly surfacing and multi-turn collaborative workflows where humans and agents share decision-making. Designing these patterns well requires methods that go beyond standard UX practice.

What is the difference between an agentic experience and a chatbot?

A chatbot is a reactive system that responds to prompts within a single interaction. An agentic system operates with the goal of planning a sequence of actions, connect to external tools and data sources, act within live workflows and reflect on outcomes to refine its next step. Where a chatbot answers a question, an agent can complete a process. The design implications are fundamentally different because the agent is a participant in the experience, not a layer on top of it.

Does the methodology only address the front-facing agent the user sees?

No. The visible agent is often an orchestrator that coordinates a network of specialised agents operating in the background. Those back-end agents access data, trigger actions in connected systems and pass results between one another before anything reaches the user. The Spark Reply methodology covers the full orchestration architecture, from the interaction layer through to the underlying agent topology and system integrations. Treating the front end in isolation typically produces results that are little better than a more polished chatbot.

What is an AX Strategy Blueprint and what does it contain?

The AX Strategy Blueprint is the primary output of the Framing and Scoping phase. It documents the business objectives the agent must serve, the user intents the agent will address, ranked by priority and feasibility, the autonomy boundaries that define what the agent can decide and act on without human confirmation, the collaboration model between humans and the agent at each point in the journey, and the integrations with tools and workflows the agent requires. Agents are represented as named actors within the blueprint, with defined roles, decision rights and escalation paths.

How can you ensure users maintain control when delegating tasks to autonomous agents?

Agentic systems inherently redistribute decision-making authority. To prevent users from losing trust or feeling overwhelmed, the experience must feature clearly defined autonomy boundaries and collaboration models. This involves designing specific interaction patterns for each stage of a journey, explicitly detailing whether the agent suggests an action, recommends a path, or acts autonomously. Crucially, it also requires integrating human-in-the-loop moments and confirmation steps, so the user retains ultimate oversight of critical decisions.

How is risk managed in autonomous agent design?

Governance and risk are managed by structurally embedding control boundaries prior to development. Guardrails, permissions, auditability, and regulatory compliance are integrated into the initial design rather than appended after deployment. Additionally, recovery mechanisms, including fallback behaviours and human-in-the-loop escalation paths, are treated as primary design requirements to ensure safe operation.

Can agentic experiences be deployed across multiple channels?

How is the success of an agentic experience measured after launch?

Standard user experience metrics such as task completion rate and time on task are insufficient for agentic systems. AX-specific KPIs are defined during the Testing and Refining phase before deployment. These metrics are then tracked continuously post-launch and cover dimensions including intent resolution accuracy, autonomous action rate versus human intervention rate, escalation frequency and outcomes, edge case handling quality and user-perceived trust in the agent. This data feeds directly into the continuous improvement cycle.

Leverage Spark Reply's Unique Expertise

Spark Reply specialises in the intersection of human experience and emerging technology. The AX approach addresses the full stack of agentic design, from strategic framing through interaction design, behavioural specification, testing and post-deployment monitoring. The methodology applies whether an organisation is deploying a single task-execution agent or a multi-agent orchestration system spanning multiple business processes across Employee Experience and Customer Experience domains.

Spark Reply

Spark Reply, a company of Reply Group, is a pioneer in designing desirable product-service experiences. Through its fast business prototyping method, Spark Reply accelerates entrepreneurship and culture of innovation within companies. Spark Reply combines talent across business, technology and design to bring physical and digital innovations to life and solve even the toughest growth challenges.

Designing Agentic Experiences

Boosting the Adoption of Agentic Systems