model interface design

This study addresses the lack of systematic investigation into architectural design decisions for non-large language model components in current AI agent systems. The authors propose a protocol-guided, source code–driven empirical analysis method that enables, for the first time, transparent deconstruction of heterogeneous AI agent systems. Through cross-project qualitative coding and co-occurrence analysis of 70 open-source projects, they identify five core design dimensions—sub-agent architecture, context management, tooling systems, security mechanisms, and orchestration—and uncover their combinatorial patterns. Based on these findings, the study further distills five archetypal architectural patterns: lightweight tool-oriented, CLI framework–based, multi-agent orchestrator, enterprise system, and domain-specific vertical architectures.

Current AI coding agents, in the absence of explicit architectural design, rapidly make unvetted software architecture decisions through implicit mechanisms. This work identifies five such implicit architecture-shaping mechanisms and introduces, for the first time, the concept of “vibe architecting,” establishing a mapping between natural language prompt characteristics and system architectural requirements. It further distills six prompt-architecture coupling patterns. Integrating large language model prompt engineering, tool-call orchestration, structured output validation, and architectural decision recording, the study empirically demonstrates that subtle differences in prompt phrasing alone can yield substantially divergent system architectures. Building on these findings, the paper proposes corresponding review practices and tooling support to advocate for effective governance of implicit architectural decisions.

This work addresses the lack of formal rules in UML modeling, which often leads to semantic inconsistencies across diagrams and compromises architectural integrity. To resolve this, the paper proposes composite consistency rules that, for the first time, formally encode architects’ design practices into composable and reusable high-level patterns. These patterns enable the automatic derivation of target diagrams through abstraction via natural-language-based rules. The approach has been implemented as an automated JScript extension within Sparx Enterprise Architect, significantly improving modeling consistency and completeness while reducing redundant manual operations. This automation accelerates the design process and enhances the reusability of UML architectures across multiple projects, thereby laying a foundational framework for AI-assisted architectural generation.

This work addresses the lack of systematic architectural approaches for enterprise-scale multi-agent collaborative systems, particularly in complex scenarios integrating human and AI agents. The authors propose a three-layer design pattern—comprising LLM agents, autonomous agents, and agent communities—that integrates principles from distributed coordination, formal modeling, and a role-protocol-governance structure. For the first time, this framework introduces formal collaboration protocols and role-based governance mechanisms into agent communities, enabling executable specification and verification of organizational, legal, and ethical rules. Validated through a clinical trial matching case study, the architecture demonstrates governable and verifiable human-AI collaboration, offering both formal verification capabilities and actionable design guidance for enterprise deployment.

This study addresses the risks associated with ungoverned conversational AI deployments in enterprise settings, which often result in low reliability, architectural degradation, security vulnerabilities, and technical debt. To mitigate these challenges, the paper introduces a Specification-Driven Development (SDD) paradigm anchored by a novel Specification Governance Reference Model (SGRM). This model enforces probabilistic AI outputs through specification contracts, a three-tier rigor framework, and deterministic verification mechanisms, thereby transforming generative AI into auditable engineering practice. The approach integrates constitutional constraints, mappings to the ISO/IEC 25010 quality model, and an agent-driven delivery pipeline. Empirical evaluation demonstrates that the proposed framework simultaneously ensures regulatory compliance and safety while reducing security defects by 73% and accelerating time-to-market by 50%.

This study addresses the prevailing gap in AI education, which emphasizes model development while neglecting system engineering practices, leaving students ill-equipped to handle real-world challenges such as architectural design, deployment, and monitoring. To bridge this gap, the authors implemented a master’s-level course in which students built a movie recommendation system under realistic constraints, with a focus on integrating AI components into robust software systems, adopting data-driven machine learning practices, and cultivating systems-level thinking. Using a mixed-methods approach—combining analysis of student project artifacts with survey data—the research evaluates learners’ performance in architectural decision-making, integration of heterogeneous models, and adaptation to evolving requirements. Findings reveal common difficulties students encounter in AI system engineering and demonstrate the course’s effectiveness in addressing critical deficiencies in AI engineering education and enhancing systems-aware competencies.