CAIN 2027
Dates to be announced Dublin, Ireland
co-located with ICSE 2027
ICSE 2027 (series) / CAIN 2027 (series) /

CAIN Scope

Software Engineering for AI-Enabled Systems/AI Engineering

The scope of CAIN is Software Engineering for AI-Enabled Systems, or AI Engineering for short, that is, the use and innovation of software engineering principles and techniques to build software systems that use one or more AI/ML models. CAIN papers often apply, expand, adapt, or invent software engineering principles to the engineering of AI-enabled systems. CAIN is interested in the engineering of all kinds of AI-enabled systems, including end-user software (e.g., personal assistants, web applications, customer support agents), developer tools (e.g., agentic coding, IDE extensions), and components with substantial non-ML engineering parts (e.g., scalable ML deployment infrastructure, compound AI systems, API design and documentation, RAG).

CAIN strives to be a venue for deep discussions, proposals, and solutions at the intersection of software engineering and AI. While there are many well-established venues in the fields of AI and machine learning, CAIN is unique in that it takes a systems and engineering perspective, considering the entire system and not just the model. CAIN focuses on building larger components or entire applications, not on improving the accuracy of individual models on benchmarks. CAIN is interested in requirements, integration testing, system-level assurance, architecture, process, and many other software engineering topics as applied to AI-enabled systems. Topics like fairness, explainability, and security are clearly in scope if they consider the system around the model, its interaction with the environment, or the interactions of multiple models. All submissions MUST have a clear connection to software engineering, a system perspective, or a life cycle perspective.

For CAIN 2027, we are looking in particular for contributions in the following areas:

  • Engineering principles, architecture, and quality assurance for agentic software, especially beyond coding agents
  • Requirements engineering for systems with AI components (e.g., system vs model requirements)
  • System-level security and safety engineering strategies for systems with AI components

The following topics are explicitly NOT in scope and will be desk-rejected:

  • Entirely model-centric papers: ML learning algorithms and representations, model optimization, and model benchmarking without a clear connection to the system that uses the model are not in scope. At a minimum, model-centric work must be clearly motivated by a use case of the model in a real-world application and should evaluate its contributions in this application setting.
  • Pure AI4SE papers: Papers that develop AI tools for software engineering tasks without a focus on how to build such tools or integrate such tools into a larger system are not in scope. For example, designing the harness for a coding agent is in scope (focused on the design on an AI-enabled system), a user study of explanations of an AI vulnerability detection tool is in scope (focused on the system aspect of an AI-enabled tool from a user’s perspective), but merely improving accuracy of an LLM-based classifier of GitHub issues is not (model centric work applied to an software engineering task, rather than software engineering principles applied to an AI-enabled system), and neither is the analysis of software engineering data with data science methods (model work applied to software engineering data, rather than software engineering principles applied to an AI-enabled system).

The submission form will ask explicitly for a short description of how the paper aligns with the scope of CAIN. Submissions should also make the connection to software engineering principles clear in the text of the paper. If you are still unsure whether your paper aligns with the scope of CAIN, please reach out to the program co-chairs.