Marc Scibelli
Marc Scibelli

HAX: Creating the Standards for Human Agent Collaboration

At Outshift by Cisco, I created the opensource HAX project to design the next layer of interaction between humans and intelligent agents. Through shared principles, reusable components, and an SDK, we made agent behavior visible and consistent across systems.

Challenge

Challenge

Challenge

Challenge

Challenge

Challenge

The rise of intelligent agents demands more than just “AI automates tasks.” As agents become autonomous collaborators, the interaction model shifts.

We are facing:

  • Lack of consistent behaviour models: Different teams built agents that behaved inconsistently from the user’s viewpoint, making trust and predictability difficult.

  • Limited transparency: Users often couldn’t understand why an agent did something or how much control they had.

  • Fragmented tooling: No unified component or SDK for agent-human collaboration across products.

  • Scaling human-agent relationships: As agent ecosystems grow, designing one-off workflows becomes unsustainable—users needed predictable, stable patterns.

Organization

Cisco

Role

Design Lead, Inventor

How do we design human-agent interfaces that are consistent, comprehensible and scalable across products?

Approach

Approach

Approach

Approach

We approached it with three parallel tracks:

  1. Behavioural Principles:
    We defined the core rules that any agent-human interaction should follow, regardless of domain. (e.g., humans remain in control, the agent’s reasoning is visible, accountability flows back to the user)

  2. A SDK:
    We developed an SDK that links back-end agent logic to front-end behavior and UI components, so teams can plug into the same human-agent interaction patterns rather than starting from scratch.

  3. A Component Library:
    We built a set of UI and interaction components that embed those principles in reusable form: agent notifications, hand-off flows, task orchestration panels, escalation pathways.

We treated HAX as a behavior system, not just a set of guidelines.

That meant designing for consistency and composability across diverse products and agent workflows.

A Fundamental Shift In How Humans Work

A Fundamental Shift In How Humans Work

What We Learned: Consistent patterns emerged around explainability, control, recovery, and collaboration,  shaping a shared model for how agents should communicate and act.
What We Learned: Consistent patterns emerged around explainability, control, recovery, and collaboration,  shaping a shared model for how agents should communicate and act.
What We Learned: Consistent patterns emerged around explainability, control, recovery, and collaboration,  shaping a shared model for how agents should communicate and act.

As a Result We Developed 5 Core Design Principles for Human-AI Collaboration

As a Result We Developed 5 Core Design Principles for Human-AI Collaboration

Control: Humans
guide interactions.

Establishing clear boundaries for AI Agents to ensure they operate within a well-defined scope.

Clarity: Humans
guide interactions.

Clarity: Humans
guide interactions.

Agents communicate their actions clearly, revealing just the right amount of detail through progressive disclosure so users always know what’s happening and why.

Collaboration:
Work together toward shared goals.

Agents contribute actively while remaining responsive to user preferences and feedback. The goal isn’t automation, it’s teamwork.

Recovery: Detect, explain, and help fix errors.

When something goes wrong, the agent helps the user find the issue and provides clear paths to correct it while improving next time.

Traceability: Build trust through explainability.

Traceability: Build trust through explainability.

Agents make their reasoning visible so users can see how decisions were reached, supporting transparency and accountability.

Traceability: Build trust through explainability

Control: Humans guide interactions

Establishing clear boundaries for AI Agents to ensure they operate within a well-defined scope.

Clarity: Humans guide interactions

Agents communicate their actions clearly, revealing just the right amount of detail through progressive disclosure so users always know what’s happening and why.

Collaboration: A shared goal

Agents contribute actively while remaining responsive to user preferences and feedback. The goal isn’t automation, it’s teamwork.

Recovery: Detect and explain errors

When something goes wrong, the agent helps the user find the issue and provides clear paths to correct it while improving next time.

Traceability: Trust through explainability

See agent actions in context and observe agent performance in network improvement.

The HAX Framework

The design principles gave us a foundation, but teams still needed a way to apply them consistently in real products. We set out to create a system that connects those principles to the tools, components, and checks developers use every day. That effort became HAX—a unified framework for designing, building, and governing meaningful human agent collaboration.

HAX
Principles:

Design for collaboration

Five research based, human-centered rules: Clarity, Control, Recovery, Collaboration, and Traceability that define trustworthy agent behavior.

HAX
SDK:

Build with
consistency

Toolkit that turns those principles into schemas, components, and checks so agents act and explain predictably.

Custom Repositories:

Reusable Explainability:

Behavior layer that travels with the agent. The same evidence, reasoning, and actions appear across any product or surface.

Portable Explainability:

Consistency Everywhere:

Behavior layer that travels with the agent. The same evidence, reasoning, and actions appear across any product or surface.

A unified framework for designing, building, and governing meaningful human–agent collaboration.

A unified framework for designing, building, and governing meaningful human–agent collaboration.

The HAX Framework

The design principles gave us a foundation, but teams still needed a way to apply them consistently in real products. We set out to create a system that connects those principles to the tools, components, and checks developers use every day. That effort became HAX—a unified framework for designing, building, and governing meaningful human agent collaboration.

The design principles gave us a foundation, but teams still needed a way to apply them consistently in real products. We set out to create a system that connects those principles to the tools, components, and checks developers use every day. That effort became HAX—a unified framework for designing, building, and governing meaningful human agent collaboration.

HAX
Principles:

Design for collaboration

Five research based, human-centered rules: Clarity, Control, Recovery, Collaboration, and Traceability that define trustworthy agent behavior.

HAX
SDK:

Build with
consistency

Toolkit that turns those principles into schemas, components, and checks so agents act and explain predictably.

Custom Repositories:

Reusable Explainability:

Behavior layer that travels with the agent. The same evidence, reasoning, and actions appear across any product or surface.

Portable Explainability:

Consistency Everywhere:

Behavior layer that travels with the agent. The same evidence, reasoning, and actions appear across any product or surface.

A unified framework for designing, building, and governing meaningful human–agent collaboration.

A unified framework for designing, building, and governing meaningful human–agent collaboration.

The HAX SDK

The HAX SDK

A React-based toolkit for building applications that integrate human and agent behavior. The SDK connects an app’s logic and context to interactive components, making agent interactions feel natural, transparent, and easy to monitor across interfaces.

Each Al prompt and user response would generate "cards" or "nodes"

finding the best responses.

A React-based toolkit for building applications that integrate human and agent behavior. The SDK connects an app’s logic and context to interactive components, making agent interactions feel natural, transparent, and easy to monitor across interfaces.

The HAX Component Library

The HAX Component Library

A library of modular interface elements such as Timelines, Dashboards, Visualizers, and Form Builders that bring HAX principles to life. Each component supports explainability, feedback, and collaboration between humans and agents.

These cards can then be moved,

edited, or grouped to create the best narrative result.

A library of modular interface elements such as Timelines, Dashboards, Visualizers, and Form Builders that bring HAX principles to life. Each component supports explainability, feedback, and collaboration between humans and agents.

The HAX Website

The HAX Website

A single hub for the HAX framework, connecting behavioral principles, research insights, SDK tools, and component patterns used to design intelligent agent experiences

A single hub for the HAX framework, connecting behavioral principles, research insights, SDK tools, and component patterns used to design intelligent agent experiences

HAX Highlights

HAX Highlights

Principles Framework:

Principles Framework:

Defines the five behavioral foundations: Control, Clarity, Recovery, Collaboration, and Traceability. These principles give teams a shared language for designing trustworthy and explainable systems.

Developer Toolkit

Developer Toolkit

The developer toolkit that connects agent reasoning, state, and confidence data directly to user facing components. The SDK standardizes how explainability, confidence, and control are represented across products.

Component Library

Component Library

A collection of modular UI components and design patterns that express the HAX principles in real interfaces. It includes elements for task orchestration, error recovery, confidence display, and user control.

Research-Driven

Research-Driven

Built on deep exploratory research into how people understand and collaborate with intelligent systems. The findings shaped a catalog of patterns focused on feedback, trust calibration, and orchestration.

Custom Repositories

Custom Repositories

Supports multiple repositories so teams can share, fork, and customize components or maintain private libraries with version control.

Portable Eplainability

Portable Eplainability

Extends HAX across systems so an agent’s reasoning, evidence, and confidence stay consistent wherever it appears.

How Teams Implement HAX

How Teams Implement HAX

A step-by-step look at how a team builds, shares, and customizes agent components using the HAX registry and local code integration.

A step-by-step look at how a team builds, shares, and customizes agent components using the HAX registry and local code integration.

Looking Ahead

HAX has become the foundation for how we think about human and agent collaboration across Cisco. It connects research, principles, and implementation into one evolving system that helps teams design with clarity and build with consistency.


The next phase of this work focuses on how explainability itself can travel across products and contexts. By giving agents a portable structure for reasoning, evidence, and confidence, we can maintain trust and transparency wherever they operate.


HAX is still early, but it represents a clear step toward a more connected and understandable future of intelligent systems, where people and agents collaborate with shared logic, accountability, and purpose.


HAX has become the foundation for how we think about human and agent collaboration across Cisco. It connects research, principles, and implementation into one evolving system that helps teams design with clarity and build with consistency.


The next phase of this work focuses on how explainability itself can travel across products and contexts. By giving agents a portable structure for reasoning, evidence, and confidence, we can maintain trust and transparency wherever they operate.


HAX is still early, but it represents a clear step toward a more connected and understandable future of intelligent systems, where people and agents collaborate with shared logic, accountability, and purpose.

We initiated a design discovery process which enabled a systematic exploration and validation of ideas, ensuring each phase, from discovery to release, was grounded in evidence and iterative improvement.