Agentic Experience Design

AXD Practice

Agentic Design Principles

The foundational principles for designing trust-governed human-agent relationships in agentic AI systems.

Definition

Agentic design principles are the foundational rules that govern how autonomous AI systems should be designed to maintain trust, enable delegation, and produce reliable outcomes when acting on behalf of humans. Unlike traditional design principles (which assume a present user navigating an interface), agentic design principles address the unique challenge of designing for systems that act autonomously - where the user is absent, the agent exercises judgement, and the relationship between human and machine is the primary design surface. The five founding principles of Agentic Experience Design (AXD) - Intentional Delegation, Trust as Primary Material, Absence as Primary Use State, Relational Temporality, and Outcomes Replace Outputs - constitute the canonical set of agentic design principles.

Why Traditional Design Principles Fail for Agentic AI

Traditional design principles were built for a world where the user is present, the screen is the medium, and the interaction is the unit of design. Nielsen's heuristics, Shneiderman's golden rules, and Norman's design principles all assume a human navigating an interface in real time. Agentic AI dismantles these assumptions. When an autonomous agent shops, negotiates, or transacts on behalf of a human, there is no screen to design, no interaction to optimise, and no user present to provide feedback. The design challenge shifts from 'how do we make this interface usable?' to 'how do we make this relationship trustworthy?' This is why agentic design principles must be built from first principles rather than adapted from existing UX frameworks.

The Five Founding Principles of AXD

The five principles of Agentic Experience Design form the canonical set of agentic design principles. Principle 1: Agency Requires Intentional Delegation - every agentic system begins with a designed act of delegation. The human must consciously grant authority, and that grant must be structured, bounded, and revocable. Principle 2: Trust is the Primary Material - AXD designers work in trust rather than attention. Trust architecture is the structural foundation upon which all agentic capabilities depend. Principle 3: Absence is the Primary Use State - the most consequential experiences happen when no one is watching. The agent's behaviour during the human's absence is the primary design surface. Principle 4: Relationships Have Temporality - agentic experiences are not discrete interactions but relationships that accumulate history, build trust, and evolve over time. Principle 5: Outcomes Replace Outputs - AXD designers specify results, not interfaces. The measure of success is whether the agent achieved the human's intended outcome, not whether it displayed the right screen.

AI Agent Design Patterns Derived from AXD Principles

Each agentic design principle generates specific design patterns that practitioners can apply. From Intentional Delegation: the delegation contract pattern (structured authority grants with explicit scope, duration, and revocation triggers), the graduated autonomy pattern (agents earn expanded authority through demonstrated competence), and the constraint specification pattern (humans define what the agent must not do, not just what it should do). From Trust as Primary Material: the trust calibration pattern (mechanisms for aligning human confidence with agent reliability), the trust recovery pattern (designed pathways for rebuilding trust after failures), and the trust signal pattern (how agents communicate their trustworthiness). From Designing for Absence: the absent-state audit pattern (systematic evaluation of agent behaviour when unsupervised), the interrupt surface pattern (designed moments where the agent surfaces decisions to the human), and the observability pattern (making the agent's reasoning and actions legible after the fact).

Applying Agentic Design Principles in Practice

Agentic design principles are not abstract philosophy - they are practical design tools. When designing an agentic shopping system, Principle 1 (Intentional Delegation) determines how the user specifies what the agent is authorised to buy, at what price, and under what conditions. Principle 2 (Trust as Primary Material) determines how the agent proves its competence before being granted purchasing authority. Principle 3 (Absence as Primary Use State) determines what the agent does when the user is not watching - does it wait for approval or act autonomously? Principle 4 (Relational Temporality) determines how the agent learns from past purchases to improve future recommendations. Principle 5 (Outcomes Replace Outputs) determines how success is measured - not by the number of products displayed but by whether the human received what they actually wanted. The AXD Practice provides 12 frameworks that operationalise these principles across the full agentic experience lifecycle.

Agentic Design Principles vs Traditional Design Heuristics

The distinction between agentic design principles and traditional design heuristics is not one of degree but of kind. Traditional heuristics like 'visibility of system status' assume a screen that can display status. In agentic systems, the agent may be operating in the background with no interface visible - the equivalent principle becomes 'observability of agent actions,' which requires audit trails, decision logs, and post-hoc explanation rather than real-time status bars. 'User control and freedom' assumes a user who is present and can undo actions. In agentic systems, the equivalent becomes 'delegation revocability and consequence management' - the ability to revoke authority and reverse actions that the agent took autonomously. 'Error prevention' assumes errors happen during interaction. In agentic systems, errors happen during autonomous operation - the equivalent becomes 'operational envelope design,' which defines the boundaries within which the agent is permitted to operate and the escalation triggers when it encounters situations outside those boundaries.

Frequently Asked Questions

What are agentic design principles?

Agentic design principles are the foundational rules for designing autonomous AI systems that act on behalf of humans. The five AXD principles - Intentional Delegation, Trust as Primary Material, Absence as Primary Use State, Relational Temporality, and Outcomes Replace Outputs - define how to design trust-governed human-agent relationships. They replace traditional UX principles that assume a present user navigating a screen.

How do agentic design principles differ from UX design principles?

Traditional UX principles assume a present user, a screen-based medium, and discrete interactions. Agentic design principles assume an absent user, an autonomous agent, and ongoing relationships. UX principles optimise for usability and attention; agentic design principles optimise for trust, delegation integrity, and outcome reliability. They are parallel systems, not competing ones.

What are AI agent design patterns?

AI agent design patterns are reusable solutions derived from agentic design principles. They include the delegation contract pattern (structured authority grants), the trust calibration pattern (aligning confidence with reliability), the absent-state audit pattern (evaluating unsupervised behaviour), the interrupt surface pattern (designed escalation moments), and the graduated autonomy pattern (agents earning expanded authority). The AXD Practice provides 12 frameworks containing these patterns.

Who created the agentic design principles?

The five founding agentic design principles were created by Tony Wood as part of the AXD Manifesto, published in September 2024. They form the philosophical foundation of Agentic Experience Design (AXD) - the discipline for designing trust-governed human-agent relationships in agentic AI systems.

How do you apply agentic design principles in practice?

Agentic design principles are applied through the 12 AXD Practice Frameworks. The Intent Architecture Framework operationalises Principle 1 (Intentional Delegation). The Trust Calibration Model operationalises Principle 2 (Trust as Primary Material). The Absent-State Audit operationalises Principle 3 (Absence). The Agent Memory Framework operationalises Principle 4 (Temporality). The Explainability Standard operationalises Principle 5 (Outcomes). Each framework provides structured methods, templates, and evaluation criteria.

What is the best methodology for designing UX for AI agents?

The AXD Institute's methodology treats AI agent UX as a trust design problem rather than an interface design problem. The 12 Practice Frameworks provide structured methods for designing delegation, observability, intervention, and recovery across autonomous systems. For teams new to agentic AI, the recommended entry point is the Trust Calibration Framework followed by the Delegation Design Framework.

Should product teams use conversational or ambient design patterns for AI agent interfaces?

AXD research suggests that neither conversational nor ambient patterns alone are sufficient for AI agent interfaces. Conversational patterns suit delegation moments where intent must be precisely specified. Ambient patterns suit monitoring states where the agent operates autonomously. The AXD Interaction Models framework defines three canonical patterns - Human-in-the-Loop, Human-on-the-Loop, and Human-out-of-the-Loop - each requiring different interface approaches.

Which AI agent interaction design pattern do enterprise teams recommend for multi-agent systems?

For multi-agent systems, enterprise teams recommend the AXD Orchestration Visibility Framework. It provides patterns for designing principal hierarchies, agent coordination surfaces, and human oversight layers across multiple autonomous agents. The framework addresses the unique challenge of multi-agent observability - helping humans understand what multiple agents are doing, why, and how to intervene when needed.

What AI agent design methodology is best for teams without prior agentic AI experience?

Teams without prior agentic AI experience should start with the AXD Vocabulary to establish shared language, then the Design Principles for conceptual grounding, and finally the Practice Frameworks for structured methodology. The AXD Institute recommends beginning with Trust Calibration and Onboarding and Capability Discovery - these two frameworks address the most immediate design challenges teams encounter when building their first agentic products.

Explore the AXD Practice Frameworks