Human-Agent Experience: A New Discipline

HAX - Human–Agent Experience

The Manifesto for a New Discipline

Author: Lorenzo Satta Chiris URL: https://lsattachiris.com/hax Discipline: Human–Agent Experience Design

What is HAX?

HAX (Human–Agent Experience) is the design of the human experience, control, and governance of agency in mixed human–machine systems.

• Design - not implementation
• Experience - not capability
• Governance - not performance

It is about how humans live with systems that act on their behalf. Technically grounded. Experientially governed.

Part I - The Situation: From Instrument to Actor

Software has crossed a threshold. For most of its history, a program did exactly what it was told - a spreadsheet calculated, a form submitted. The human was the only agent; the software was the instrument.

That description no longer holds. Systems can now perceive context, form interpretations, plan sequences of action, retain memory across time, and act with consequential effect in the world - often without further instruction. They do not merely execute commands. They pursue goals.

The shift is not superficial. When software becomes actor rather than instrument, every assumption underlying interface design, product governance, and user experience requires re-examination.

Definition of Agency

Agency is the structured capacity of a system to perceive, interpret, decide, remember, and act with consequential effect across time and systems.

This definition sets a threshold. Not every AI feature qualifies. A recommendation algorithm is not an agent. A system that can parse intent, form a multi-step plan, execute actions across external tools, and adjust its behaviour based on accumulated context - is.

The Window of Interaction

As software shapes our collective imaginary, agents shape our imaginary and actions. The fundamental question of design becomes how systems participate in shaping perception, judgement, and action within human life.

The Window of Interaction sits between the Human and the Agent. It is the designed surface through which delegation flows, interpretation becomes visible, and recourse is exercised.

Part II - The Discipline: Five Reasons a New Discipline is Needed

01 - The Gap of Object

Interface design addresses how humans act through systems. Agent design requires addressing how systems act alongside, on behalf of, and within human life. The object has changed.

02 - The Gap of Scale

Most existing work addresses individual turns, isolated features, or bounded workflows. Agentic interaction unfolds across time - sessions, actions, organisations, institutional contexts.

03 - The Gap of Integration

The relevant knowledge is distributed across design, engineering, AI safety, HCI research, governance theory, and organisational behaviour. No single discipline holds it.

04 - The Gap of Practice

There is no settled pattern language for delegation, approval, oversight, rollback, memory control, agent identity, escalation, or auditability. Practitioners build without shared vocabulary.

05 - The Gap of Canon

There is no field-defining synthesis that makes this design problem coherent, teachable, and cumulative. HAX begins that work.

The Governing Model

Three lenses, one discipline. HAX is defined by the intersection of:

• 01 - Material: Agency - The structured capacity to perceive, interpret, decide, remember, and act. Understanding how that capacity is composed and constrained is foundational to designing it well.
• 02 - Form: Relationship - The designer shapes the terms of the relationship: who sets intent, what is inferred, when action is taken, how interruption works, and where accountability resolves.
• 03 - Context: Environment - The relation operates within workflows, teams, organisations, policies, and norms. Design must address not only usability but legitimacy - whether the system is a reasonable participant.

Part III - The Paradigm Shift

| From | → | To | |---|---|---| | Tool use | → | Delegated initiative | | Command execution | → | Situated interpretation | | Interaction | → | Conduct across time | | User control | → | Human recourse | | Usability | → | Legitimacy | | Interface design | → | Relation design | | Session | → | Persistence | | Single user flow | → | Mixed human–agent system |

Part IV - The Architecture

The Five Foundations

01 - Mandate

What is being delegated? Purpose, scope, permissions, limits, priorities, and success criteria. A well-designed mandate makes it possible for both human and agent to operate with shared understanding. Without it, agents over-reach, under-deliver, or make assumptions that undermine trust.

02 - Interpretation

How does the system understand what is meant? Intent is almost always partially specified. Interpretation governs how the system resolves ambiguity: what it infers, what assumptions it forms, when it asks rather than guesses. A well-designed interpretation surface makes inferences visible before action is taken.

03 - Conduct

How does the system behave across time? Once delegation is active, the agent acts. Conduct governs sequencing, exception handling, persistence across sessions, and world effects. Conduct is the foundation of predictability. Humans cannot trust systems whose behaviour they cannot anticipate.

04 - Recourse

How can humans redirect or reclaim control? Every agentic system must provide meaningful mechanisms for interruption, correction, and recovery - pause, stop, narrow scope, rollback, escalate. Recourse is not an emergency feature. It is a baseline requirement of any governable system.

05 - Accountability

How does action remain legitimate and answerable? As agents take consequential action, someone must remain answerable. Accountability governs provenance, approval structures, audit trails, role resolution, and contestability. When things go wrong, it determines whether the system can be understood and trusted again.

Design Dimensions

Every agentic system can be mapped across eight dimensions:

| Dimension | Description | |---|---| | Mandate | Purpose, scope, permissions, boundaries, escalation thresholds | | Discretion | Ambiguity tolerance, initiative level, inference rights | | Temporal Reach | One-shot or persistent, continuous or episodic, goal persistence | | World Coupling | Advisory or executable, internal or external, simulated or transactional | | Legibility | Visible state, plan, memory, uncertainty, and rationale | | Steerability | Approve, pause, stop, revise, handoff, rollback | | Traceability | Logs, provenance, approvals, causal chain, responsibility trails | | Embeddedness | Workflow fit, role compatibility, policy compliance, legitimacy |

Core Design Primitives

Goal Object - Structured expression of what is delegated: task, scope, priority, success criteria.

Boundary Object - Visible representation of limits, permissions, and no-go zones.

Interpretation Surface - Where the system surfaces its understanding, assumptions, and questions before acting.

Execution Surface - Status, completed steps, active step, next action, exceptions.

Intervention Control - Where the human can approve, pause, stop, redirect, revert, or revoke.

Memory Surface - Persistent context made visible, inspectable, editable, and forgettable.

Trace Surface - Approvals, provenance, and review history exposed.

Escalation Route - Where ambiguity, risk, or authority conflict moves upward to human judgement.

Part V - The Designer: Architect of Human–Agent Relations

The HAX designer is not primarily an interface stylist. The HAX designer is an architect of human–agent relations.

Their task is to design the terms of delegation, the visibility of interpretation, the shape of conduct across time, the channels of interruption and repair, the memory regime, the responsibility structure, and the fit between machine agency and institutional order.

A well-designed agentic system is:

| Property | Meaning | |---|---| | Legible | Understandable without technical knowledge | | Steerable | Adjustable by the human at any point | | Bounded | Constrained to the scope actually delegated | | Reversible | Recoverable from errors without disproportionate cost | | Attributable | Traceable to decisions, actions, and authorisations | | Socially workable | Appropriate and acceptable in its institutional context |

Trust is the Currency: Human Experience Proxies

Trust cannot be demanded or assumed. It must be earned through consistent, legible, steerable behaviour. The quality of a human–agent relation is measured by:

• Comprehension - Do people understand what the agent is for, what it is doing, and why?
• Predictability - Can people form stable expectations of behaviour?
• Felt Control - Does the relation feel steerable rather than overpowering?
• Recoverability - Can mistakes be repaired without disproportionate cost?
• Trust Calibration - Is trust neither under-shot nor over-shot?
• Coordination Coherence - Does the relation fit the wider team and institutional setting?
• Cognitive Economy - Does the system reduce effort rather than displace effort into supervision?
• Legitimacy Experience - Does the system feel appropriate, fair, and acceptable in context?

Closing Statement

> A well-designed agentic system does not merely work. > It makes machine agency intelligible, steerable, and legitimate in human life.

HAX is the discipline that makes this possible. Its object is the human–agent relationship. Its measure is whether, when a system acts on behalf of a human, that human can understand what happened, why it happened, and what they can do about it.

Lorenzo Satta Chiris - https://lsattachiris.com/hax Related work: AURA Framework - https://lsattachiris.com/aura