Track Co-Chairs
Johnny Chan, University of Auckland
Adnan Mahmood, Macquarie University
Michael Sheng, Macquarie University
Kasuni Weerasinghe, Auckland University of Technology
Track Description
The expanding capabilities of Generative AI are redefining software development at a pace that outstrips existing theoretical and empirical understanding. This track aims to provide a forum for examining this rapidly evolving landscape, from the emergence of vibe coding and agentic AI systems through to the latest practice of harness engineering, highlighting both the technological advancements and the broader implications of these emergent technologies for information systems research.
The field has seen a rapid accumulation of new paradigms, each building on its predecessors. In early 2025, vibe coding popularised AI-assisted development in which developers describe intent in natural language and accept generated code with minimal manual intervention. Agentic AI systems extended this by introducing tool-using agents capable of planning, executing and verifying multi-step work across repositories, tests and deployment pipelines, often producing reviewable outputs rather than isolated code snippets. Most recently, industry has demonstrated that building an agent is comparatively straightforward; making it reliable, safe and maintainable in production is where the real challenge lies. This recognition has given rise to harness engineering, an emerging set of practices concerned with designing the constraints, feedback loops, documentation and verification systems that wrap around AI agents to make them dependable at scale. The engineer’s primary job shifts from writing code to designing environments in which agents can do reliable work. Among these paradigms, vibe coding remains the fastest path for prototyping and exploration; agentic workflows drive production-grade implementation; and harness engineering provides the scaffolding that makes agentic work trustworthy at scale. Taken together, they reframe software development as human–agent collaboration, in which engineers increasingly shape the conditions and specify intent while AI systems implement, test and iterate.
We seek to attract research that delineates, critiques and extends the concepts, methods, frameworks, architectures and broader implications of applying Generative AI, agentic workflows and harness engineering in IS design and software development.
The scope of this track includes, but is not limited to, the following key areas:
- Harness engineering as an emerging practice: the design of constraints, tooling, feedback loops, documentation and verification systems that govern AI agent behaviour. This includes empirical studies of harness components, investigations of how harness maturity relates to agent reliability, and studies of the evolving role of the software engineer from code author to environment architect.
- Agentic AI in software development: the design, deployment and governance of autonomous coding agents that plan, implement, test and submit code with limited human intervention. Topics include multi-agent orchestration, agent memory and context management, interoperability standards (e.g. Model Context Protocol, Agent-to-Agent Protocol), and the socio-technical implications of delegating implementation decisions to AI systems.
- Autonomous AI experimentation and self-improving systems: agent-driven research loops in which AI systems autonomously propose, execute, evaluate and iterate on experiments without human intervention (e.g. AutoResearch). Research topics include the design of constrained experimentation harnesses, the shift from researcher-as-coder to researcher-as-strategist, and evaluation bottlenecks when experiment throughput exceeds human review capacity.
- The evolution of AI-assisted development paradigms: from vibe coding (i.e. natural-language intent and minimal oversight) through specification-driven development and context engineering to harness engineering, including comparative studies of these paradigms and their suitability across project types, team structures and risk profiles.
- Organisational adoption, strategy and value creation: how organisations evaluate, adopt and govern these emerging technologies, including technology selection and implementation strategies, alignment with business objectives, the changing relationship between IT and business units, and implications for digital transformation when AI agents become active participants in development workflows.
- The changing nature of developer work and expertise: how the shift from writing code to designing agent environments affects skill development, professional identity, team composition and organisational capability. This includes studies of deskilling and reskilling, the emergence of new roles (e.g. harness engineer, agent orchestrator), and implications for IS education and workforce development.
- Ethical, governance and accountability dimensions: intellectual property rights and attribution in agent-generated outputs, bias and provenance tracking, accountability when agents act across tools and environments, responsible AI governance structures for agentic development, and the broader societal implications of democratising software creation through AI.
- Productivity, quality and risk trade-offs: the tension between accelerated development and downstream consequences, including security vulnerabilities in AI-generated code, compliance and privacy concerns, automation bias, organisational over-reliance on opaque agent systems, and the strain on review capacity and delivery infrastructure when code production outpaces human oversight.
This track aspires to be a platform for rigorous scholarly inquiry into the multifaceted applications of Generative AI in software development, emphasising both the innovative potential and the consequential ethical, legal and operational challenges. We welcome empirical studies, design science research, theory-building, field experiments and mixed-method work grounded in real development settings.