Everything Evolves, Something Endures: The 1st Workshop on Harness Engineering for Generative UIHarness4GenUI 2026

Generative UI (GenUI) is emerging as a new paradigm in which user interfaces are not statically designed, but dynamically generated by AI models based on intent, context, and interaction. This shift transforms UI from an artefact into a runtime system, fundamentally changing how UI systems are specified, constructed, rendered, and evolved. While advances in large language models and multimodal systems have enabled powerful generative capabilities, the engineering foundations required to build reliable, and maintainable GenUI systems remain underexplored.

Rather than treating Generative UI as purely dynamic, this workshop focuses on identifying the stable core that underpins GenUI systems and the dynamic parts that support flexibility while retaining reliability. To this end, we introduce harness engineering as a central concept, capturing the orchestration layer that governs how context, tools, data, and rendering pipelines are coordinated. Harness engineering provides structure, control, and guarantees over inherently adaptive and non-deterministic components. Building on this, we explore how software engineering principles can offer enduring structures, abstractions, and constraints within highly adaptive and AI-driven UI generation processes, including:

• Defining stable abstractions (e.g., context models, UI schemas, and interaction protocols);
• Establishing contracts and constraints across generation, orchestration, rendering and interaction.
• Separating stable system backbones from adaptive, model-driven components.
• Enabling system evolution while preserving key qualities such as maintainability and consistency.

By framing Generative UI as a software system rather than a design artefact, this workshop brings together researchers and practitioners to explore the software engineering foundations of Generative UI systems. It highlights an end-to-end perspective spanning context engineering, harness engineering, rendering pipelines, protocol design, and human–AI interaction models, aiming to establish principled approaches for building next-generation AI-native interfaces.

Accepted Papers

Call for Enduring Foundations

Generative UI is reshaping how user interfaces are conceived and constructed, moving from static artefacts to dynamically generated, context-aware systems. This transformation introduces new challenges for software engineering, including how to design, implement, and operate GenUI systems that are reliable and maintainable.

Our Harness4GenUI workshop invites researchers and practitioners to submit original work on the engineering foundations of Generative UI systems. We particularly encourage contributions that treat GenUI as a system pipeline and explore the stable core underlying dynamic UI generation.

Topics of Interest

Topics include, but are not limited to:

• Context engineering for Generative UI (context models, memory systems, retrieval pipelines)
• Harness engineering and infrastructure for GenUI systems (orchestration, runtime environments)
• DevOps/DataOps for UI generation pipelines
• GenUI design systems, UI schemas, and component abstractions
• Rendering pipelines for generative interfaces
• Protocols and contracts for UI generation and interaction
• Agent-based UI design and human–AI interaction
• Usability and deceptive patterns
• Testing, validation, and evaluation of Generative UI systems
• Observability, monitoring, and debugging of GenUI pipelines
• Trustworthiness, safety, and governance in GenUI systems

Submission Guidelines

All papers must be original and must not have been previously published or be under review for any other publication venue. Submissions must follow the formatting and submission guidelines of the main conference: ASE 2026 - Research Papers - ASE 2026

We accept three types of submissions:

• Short papers (up to 4 pages, including references)
• Industry Showcase (up to 2 pages, including references)
• Regular papers (up to 8 pages, including references)

No additional pages are permitted.

The workshop follows a single-blind review process.

Robert Mao

Founder & CEO @ ArcBlock | Blockchain & DID Pioneer | Former Microsoft Researcher | Author & Public Speaker

Keynote Title
What Endures When Everything Evolves? AFS-UI and the Search for Stable Foundations in Generative UI

Abstract

Generative UI is rapidly transforming user interfaces from static artifacts into dynamic systems that are continuously assembled, adapted, and evolved by AI. As models, rendering technologies, interaction patterns, and data sources become increasingly fluid, an important software engineering question emerges: what should remain stable?

In this talk, I will discuss lessons learned from designing AFS-UI, a framework that treats user interfaces not as generated artifacts, but as projections over a shared operational context. Rather than focusing on specific models, components, or rendering technologies, AFS-UI explores a set of enduring foundations for Generative UI systems, including shared context, uniform addressing, stable operational semantics, separation of context and projection, and system-wide composability.

I will argue that many current Generative UI approaches focus on what evolves while paying less attention to what must endure. By examining Generative UI through the lens of software architecture and systems design, we can identify stable abstractions that allow interfaces, data sources, agents, and rendering environments to evolve independently while preserving maintainability, interoperability, and long-term evolvability.

The presentation itself will be delivered as an AFS-UI application, providing a live demonstration of the ideas discussed throughout the talk.

Bio

Robert Mao is the Founder and CEO of ArcBlock, a platform that empowers developers to build decentralized applications, blockchain systems, decentralized identity (DID), and large language model (LLM)-powered tools.

A serial entrepreneur and former researcher at Microsoft Research, Robert brings deep expertise in social computing, blockchain architecture, and digital identity. He is widely recognized for his thought leadership in decentralized systems and has published over 30 papers, journal articles, and patents in the field.

Robert is also the author of two books on blockchain technology and a frequent speaker at global conferences on blockchain , identity, and emerging technologies.

Philippe Palanque

Professor at Université de Toulouse

Keynote Title
GenUI and User Centered Design and their Relationship with Automation Design and Engineering of Interactive Systems and Technologies

Abstract

User Centered Design is a generic process for the design of usable interactive applications. GenUI principles target at automating, at least, part of this process. What has been stable over the last 50 years in that domain, is the knowledge about such automations of human-driven processes and the importance of the human involvement.

Automation, as a design goal, focusses mainly on the migration of tasks from a human operator to a mechanical, chemical or digital technology. Designing automation thus usually consists in removing tasks or activities from that operator and in designing technologies that will perform them. When these automation technologies are not adequately designed (or correctly understood by the operator), they may result in so called automation surprises that degrade, instead of enhance, the overall performance of the couple (operator, system).

This presentation proposes a decomposition of automation for interactive systems highlighting the diverse objectives it may target at. Beyond, multiple complementary views of automation for interactive systems design are presented to better define the multiform concept of automation within this specific context. It provides numerous concrete examples illustrating each view from different application domains including automotive, aviation or air traffic control. The deployment of AI technologies in multiple areas maps perfectly with the initial concept of automation. However, automation fundamental guidelines and principles are usually not considered or violated when these AI technologies are designed and deployed. Depending on where automation takes place with respect to human behavior (motor, perceptive of cognitive levels) the benefits, limitations and drawbacks of automation are fundamentally different. In the context of the presentation, the dual automation concern will be explicitly addressed: automation of designers and engineers work and automation of users’ work by user interfaces automating “part” of their work.

Bio

Philippe Palanque is professor in Computer Science at the University Toulouse 3 and is head of the Reliability of Systems and Software research department at IRIT. Since the early 90’s his research focus is on interactive systems engineering proposing notations, methods and tools to integrate multiple properties such as usability, dependability, resilience and more recently user experience. These contributions have been developed together with industrial partners from various application domains such as civil aviation, air traffic management or satellite and launchers ground segments. He has been involved in the specification of future interactive cockpits of civil aircraft including their interactions means and in the modelling of operational states of civil aircraft (with direct support from and close collaboration with Airbus). He has been working in the area of automation for more than ten years, was a member of the SESAR Higher Automation Levels in Aviation network of excellence and paper co-chair of ATACCS (Application and Theory of Automation in Command and Control Systems) 2015 conference.

He was steering committee chair of the CHI conference series at ACM SIGCHI, is a member of the CHI academy and previously chair of IFIP Technical Committee on Human-Computer Interaction (TC13). He edited and co-edited more than twenty books or conference proceedings including the “Handbook on Formal Methods in Human-Computer Interaction” published by Springer in 2017.

9-10am: Connect & Set the Scene

Introduction & Icebreaker (15 min)

• Interactive activity to get to know each other and share expectations for the workshop

Academic Keynote (45 min)

• GenUI and User Centered Design and their Relationship with Automation Design and Engineering of Interactive Systems and Technologies by Philippe Palanque

10.30am-12.30pm: Explore Emerging Research and Practice

Industry Keynote (45 min)

• What Endures When Everything Evolves? AFS-UI and the Search for Stable Foundations in Generative UI by Robert Mao

Paper Presentations (60 min)

1. Toward Frontier-Quality Declarative UI Generation at Small-Model Cost (20 min)
2. The Driver Was Never the Problem: Cross-Platform Test Portability as a Semantic Contract (20 min)
3. Prompt-to-Feature Attribution for Explainable Software Generation (20 min, short paper)

Interactive Activity (15 min)

• An interactive activity to transition from the keynote into the final community discussion.

• Guiding question: What should remain stable or endure in software development as AI and generative technologies continue to evolve

2-4pm: Shape What Comes Next

Demos (40 min)

• Mini presentations/demos (5 min each)

Panel Discussion & Round Table (60 min)

• Panellists: Workshop organisers and keynote speakers

• Discussion questions:

  • What aspects of software engineering are likely to remain fundamentally human or require human-centred design?
  • Which current software development practices will need to change because of generative AI?
  • What foundations or principles should guide the design and engineering of generative UI and AI-assisted software systems?
  • What are the key open research and industry challenges that the community should work on next?

Closing (20 min)

• Reflections and key takeaways
• Looking ahead to the next edition of “Everything Evolves, Something Endures: Workshop on Harness Engineering for Generative UI”

Optional Self-Paid Dinner

Continue the conversation informally with workshop participants and organisers after the workshop.