Next-Gen Agentic AI in UX Design: Evolving the Double-Diamond Process

Multi-Agent Orchestration

A single hyper-agent rarely completes a complex task. Instead, agentic AI relies on multi-agent systems (MAS) that function similarly to a corporate team. An orchestrator agent receives a prompt to execute a human user’s goal and breaks down the prompt, then delegates tasks to specialized agents. For example, in software development, these agents might serve the following roles:

• planner agent—This agent reads a user’s feature request and creates a roadmap for its development.
• coder agent—This agent writes the actual code that implements the feature.
• reviewer agent—This agent tests the code for bugs and security flaws.

These agents share common memory, collaborate, and pass results back to the orchestrator to deliver the final product.

Agentic Al systems represent a shift in UX design, in which the Al not only responds to user inputs but also independently plans, acts, observes, and adapts. The agents’ autonomy introduces new considerations for UX and user interface (UI) design, particularly building trust, ensuring control, and facilitating ongoing interactions.

Comparing the Double-Diamond Process with Agentic UX

The traditional double-diamond design process comprises four phases: Discover, Define, Develop, and Deliver, as shown in Figure 1. It assumes a linear, human-driven process that is constrained by how fast the human brain can synthesize data. In contrast, the agentic double diamond, which is often conceptualized as the AURA (Automate, UX, Review, Approve) Framework, shifts the role of the UX designer from the sole creator to a strategic director.

Table 1 summarizes the key differences between the traditional and agentic UX double-diamond design processes.

Integrating Agentic AI Design into Existing UX Design Methods

To incorporate agentic Al into the double-diamond design process, adapt each phase to account for autonomy and loops. An integrated workflow follows, using an auto-scheduler as an example.

Discover: Agent-First Research

Focus on identifying delegable tasks and data needs. Conduct contextual interviews and diary studies to map current friction such as manual scheduling. Deliverables include a delegation audit, data inventory—for example, of calendars and email messages—and a risk register for privacy and ethics. This phase uncovers boundary errors that agents would amplify.

Define: Policy and Autonomy Matrix

Define scope, roles, and autonomy levels. Create an autonomy matrix with use cases as rows and levels as columns—for example, suggest-only, assistive, semi-autonomous, or fully autonomous. Include metrics such as time saved and rollback rates. This phase aligns UX design, legal, machine learning, and product teams on guardrails. Figure 2 illustrates the autonomy spectrum for Al agents—from rules-based to adaptive agency.

Develop: Agent Loops and Human-in-the-Loop

Build the core agent loop (Plan-Act-Observe) with stop conditions and fallbacks. Design user interfaces to show intent, steps, and edits, surfacing confidence and provenance. Define human gates for high-risk actions such as cross-organization invitations. Use tools such as LangChain for hooks and Promptlayer for monitoring. Prototyping and Wizard-of-Oz testing are useful for trust validation.

The Intent-Action-Audit (IAA) Design Framework

When developing agentic AI applications, you cannot rely on traditional navigation flows. Instead, build the user experience around the IAA Framework, as follows:

1. Intent Preview—Before taking an irreversible action, the agent summarizes its understanding of the goal and its proposed plan.
2. Autonomous Action—The agent executes the plan in the background, minimizing the need for the user to watch every step.
3. Audit & Verification—The system provides a clear, digestible log of what the agent has done, why it was done, and an easy path to undo or override the action.

Design Principles for Agentic AI

Developing an agentic AI requires a new set of UX design heuristics, as follows:

• transparency of thought—Agents must explain their rationale. For example, if an agent books a specific flight, it should state: “I chose this flight because it aligns with the user’s preference for morning departures, despite their costing $20 more.”
• variable autonomy (The Autonomy Dial)—Users need control over how much the agent can do independently. Allow users to toggle between “Ask me before taking action” and “Execute automatically and send a summary.”
• graceful escalation (Human-in-the-Loop)—When an agent encounters ambiguity or a high-risk decision, it must seamlessly pause and escalate to the human supervisor, providing the full context rather than guessing.
• intervention over navigation—Traditional UX design might use Next and Back buttons. In contrast, agentic UX design would prioritize Interrupt, Correct, and Undo buttons, allowing users to steer the agent mid-task.

Core Agentic UX Patterns

Some core agentic UX patterns include the following:

• The Progress Ledger—This is a real-time, collapsible timeline showing what the agent is currently doing—for example, Thinking → Searching database → Drafting email message → Waiting for approval.
• Confidence Signals—Visual indicators such as a color scale or percentage show how certain the agent is about its proposed action, prompting the human user to review low-confidence tasks more closely.
• The Sandbox Preview—A safe environment in which the agent simulates the outcome of an action such as executing a complex database migration before the user clicks Approve.

Standard Guidelines for Agentic UX

Several emerging standards guide agentic design, as follows:

• Google’s People + Al Guidebook—Emphasizes human-centered Al design with patterns for consent, explanations, and mental models. These are especially relevant for user trust and complexity in autonomous systems.
• ReAct Patterns—Interleaves reasoning and actions for explainable loops, enhancing UX design transparency.
• Anthropic’s Model Context Protocol (MCP)—Standardizes data connections for agents, supporting governance and integration.
• Microsoft’s Agent UX Principles—Focuses on embracing uncertainty, transparency, and user control across past, present, and future interactions.
• Salesforce Agentic Interface Guidelines—Provides tips such as codifying learnings for trust and workflows.

Conclusion

Agentic Al is not a feature we add to existing products. It represents a paradigm shift in how humans and software interact. The passive, click-driven user interface is giving way to goal-directed, conversational, autonomous experiences in which Al agents act as genuine collaborators within human workflows.

For UX designers, this shift demands new skills, new principles, and new frameworks. The double diamond remains a powerful mental model for UX design, but it must evolve by adding the Adapt phase; thus, integrating Al partnership into every stage of the design process and centering on transparency, trust, and meaningful human control as design pillars.

The opportunity that agentic AI presents is extraordinary: UX designers who master the design of agentic AI user experiences have the chance to create experiences that are not just usable or delightful, but genuinely transformative. Al partners can understand users’ goals, amplify their capabilities, and complete complex tasks with the competence of an expert collaborator and the loyalty of a trusted friend.