Abstract: Generative Artificial Intelligence (GenAI) has emerged as one of the most transformative technologies in the field of software engineering. Its ability to automate code generation, documentation, testing, and architectural design is redefining traditional workflows across the Software Development Lifecycle (SDLC). This study explores the qualitative impact of GenAI on development processes, team dynamics, productivity, quality assurance, and decision-making. Through thematic analysis of existing literature and expert perspectives, the research identifies how GenAI accelerates development while simultaneously introducing new challenges related to trust, model reliability, ethical use, and skill dependencies. The findings show that GenAI acts as a powerful collaborator by automating repetitive and time-intensive tasks, enhancing developer efficiency, and supporting high-level decision-making. However, AI-generated outputs may contain inaccuracies, introduce security vulnerabilities, or lead to over-reliance among developers. Concerns also arise regarding intellectual property, data privacy, and responsible AI governance. This paper concludes that GenAI will not replace developers but will redefine the nature of software engineering through hybrid human–AI collaboration. The study also highlights the necessity of developing regulatory frameworks, AI literacy programs, and transparent evaluation systems to ensure the safe and effective future integration of GenAI in SDLC. Keywords: Generative AI, Software Development Lifecycle, AI-Assisted Programming, Automation, LLMs in Software Engineering, Code Generation.

Large language model (LLM)-based artificial intelligence (AI) agents are powerful tools that can help researchers automate complex tasks such as literature review, data mining, computational code generation, and summarization of existing knowledge, but they can still fall short in developing original biological hypotheses and insights (see related Research Article by Rodriguez-Coffinet etal. in this issue). Emerging advances in multiagent systems and human-agent collaborative frameworks offer promising steps forward.

Deploying Large Language Models (LLMs) on resource-constrained devices remains challenging due to limited memory, lack of GPUs, and the complexity of existing runtimes. In this paper, we introduce TranSQL + , a template-based code generator that translates LLM computation graphs into pure SQL queries for execution in relational databases. Without relying on external libraries, TranSQL + , leverages mature database features-such as vectorized execution and out-of-core processing-for efficient inference. We further propose a row-to-column (ROW2COL) optimization that improves join efficiency in matrix operations. Evaluated on Llama3-8B and DeepSeekMoE models, TranSQL + achieves up to 20× lower prefill latency and 4× higher decoding speed compared to DeepSpeed Inference and Llama.cpp in low-memory and CPU-only configurations. Our results highlight relational databases as a practical environment for LLMs on low-resource hardware.

Experts Stealthy Processor Exploitation and Concealment Through Reconfigurable Elements (SPECTRE) is the new framework proposed in this paper to use the computational methods of Natural Language Processing (NLP) to automate the addition of Hardware Trojans (HTs) to the complex hardware design. SPECTRE takes advantage of Large Language Models (LLMs) like Generative Pre-trained Transformer (GPT)-4, Gemini-1.5-pro and LLaMA-3-70B to synthesize HTs with little to no human input by using sophisticated prompting methods like role-based prompting, reflexive validation prompting, and contextual Trojan prompting to analyze Hardware Description Language (HDL) codebases and expose vulnerabilities. Such a methodology alleviates the limitations of the more traditional machine learning-based automation that tends to require large data and extensive training times by including NLP-based code generation and an inference engine that can dynamically scale to non-homogeneous hardware platforms, including Application-Specific Integrated Circuits (ASICs) and Field-Programmable Gate Arrays (FPGAs). When tested on benchmark hardware systems like Static Random-Access Memory (SRAM), Advanced Encryption Standard (AES-128), and Universal Asynchronous Receiver-Transmitter (UART), SPECTRE shows greater performance with GPT-4 having an 88.88% success rate in generating viable and stealthy HTs that cannot be detected by current state-of-the-art Machine Learning (ML)-based tools like hw2vec. The mathematical and computational basis of the framework, premised on few-shot learning, adversarial prompting, and iterative validation algorithms, shows the dual-use potential of NLP models in the domain of hardware security, which poses the potential to exploit vulnerabilities within a short time but also demands adequate strategies to curb vulnerability exploitation by artificial intelligence generation.

MLSToolbox Code Generator: A tool for generating quality ML pipelines for ML systems

Performance and interpretability analysis of code generation large language models

Code like an economist: Analyzing LLMs’ code generation capabilities in economics and finance

Binding of C++ and JavaScript through automated glue code generation

Agentic AI systems have achieved wide adoption in software development, assisting engineers with code generation, diagnostics, and efficiency. However, the evolution of AI-powered agents for support engineers and IT operations remains nascent. This paper surveys the landscape of agentic AI in both software development and operational support, examines current capabilities such as observability, root cause analysis, and explores the potential for automated remediation, knowledge-based case resolution, and SME-guided learning. We propose a taxonomy of support-centric AI agent tasks and discuss research and engineering challenges for autonomous, end-to-end product support by AI.

FusionIDE is an innovative, AI-driven Integrated Development Environment (IDE) developed to transform how developers collaborate, code, and automate software development tasks. The primary objective of this project is to create a unified platform that integrates artificial intelligence and real-time collaboration to enhance developer productivity, reduce coding effort, and improve overall software quality. Traditional IDEs are limited by single-user workflows, fragmented toolchains, and lack of intelligent assistance, often resulting in reduced efficiency and coordination challenges in distributed teams. To address these issues, FusionIDE introduces a cloud-based collaborative IDE that allows multiple developers to edit, review, and debug code simultaneously with live synchronization and conflict-free editing. The system incorporates an AI pair programmer capable of generating code snippets, detecting and correcting errors, and providing contextual explanations. Additionally, voice-to-code functionality enables natural language-based programming, while the automatic UML generator produces design diagrams directly from code or textual requirements. The system is implemented using a MERN-based stack with integrated OpenAI APIs for intelligent assistance and GitHub APIs for version control. Experimental evaluation demonstrated stable real-time collaboration with minimal latency and high accuracy in voice-based code generation. The results confirm that FusionIDE significantly enhances team coordination, reduces manual effort, and streamlines the software development process, representing a step forward in intelligent, collaborative software engineering.

Programmable Logic Controllers (PLCs) are fundamental to industrial automation systems. However, traditional PLC programming requires extensive domain expertise and significant time investment, while code reusability remains limited and cross-platform adaptation poses substantial challenges. With the rapid advancement of Large Language Models (LLMs), LLM-based code generation offers a promising approach to address these issues. Nevertheless, existing methods still face challenges when handling complex industrial scenarios, including insufficient domain knowledge, unstable code quality, and weak cross-platform adaptation capabilities. This paper presents a multi-agent system for intelligent cross-platform PLC code generation, featuring a collaborative framework consisting of four specialized agents: requirement analysis, architecture design, code generation, and verification-optimization. The method injects domain knowledge through a Retrieval-Augmented Generation (RAG) mechanism, employs multi-stage prompt engineering strategies to guide code generation, and integrates a three-layer verification mechanism comprising static analysis, dynamic simulation, and expert review to ensure code quality. Experiments on the constructed PLC-MultiTask dataset demonstrate that our method significantly outperforms existing approaches across multiple metrics, achieving 90.3% compilation success rate, 87.6% test pass rate, and 75.4 CodeBLEU score. In an industrial case study involving robotic arm handling of refractory bricks, the system successfully generated approximately 800 lines of structured text code. Field testing over 720 hours demonstrated stable operation with 99.2% handling success rate, reducing development time by 73.3% compared to traditional methods. These results indicate that multi-agent-based PLC code generation significantly enhances development efficiency, ensures code quality, and strengthens cross-platform adaptation capabilities, offering a novel paradigm for industrial automation software development.

Automated code generation using large language models (LLMs) has attracted significant attention due to its potential to enhance software development. However, ensuring both accuracy and efficiency in generated code remains challenging. Prior research has mainly advanced along two directions: (i) enhancing models through architectural improvements, larger parameter scaling, and domain-specific fine-tuning; and (ii) refining prompt engineering techniques to better structure inputs and guide outputs. In this work, we pursue the latter direction and introduce a prompt engineering–based approach for Java code generation. Rather than directly generating Java code from natural language specifications, we propose a two-step pipeline: (i) generating intermediate Python code and, (ii) translating Python into Java. This design leverages the strong performance of LLMs on Python while enabling systematic optimization of the translation stage. To achieve this, we propose a set of translation strategies combining prompt engineering principles—including explicit instructions, syntax guidance, and domain keyword constraints—with advanced reasoning strategies such as Zero-shot Chain of Thought (Zero-shot-CoT) to efficiently generate Java code. Experiments on the HumanEval-X benchmark using the CodeGeeX3 model show that the proposed strategies significantly improve the accuracy of Java code generation. We further evaluate across diverse programming tasks, including file operations, HTTP APIs, database connectivity, parallel computing, and graphical applications, confirming the robustness of our approach. Finally, we validate the generality of our findings using ChatGPT (GPT-4o), observing substantial improvements over baseline prompt designs.

With the significant breakthroughs of deep learning technologies such as large language models (LLMs) in the field of code analysis, AI has evolved from an auxiliary tool to a key technology that deeply participates in code optimization and resolving performance issues. As modern software system architectures become increasingly complex, the requirements for their performance have also become more stringent. During the coding stage, developers find it difficult to effectively identify and resolve potential performance issues using traditional methods. This review focuses on the application of artificial intelligence in two key areas: AI-assisted intelligent code generation and AI-povered code review. The review systematically analyzed the application of LLMs in software development, revealing a situation where efficiency gains coexist with quality challenges. In terms of code generation, models such as Code Llama and Copilot have significantly accelerated the development process. In the field of code review, AI can effectively handle code standards and low-severity defects. However, in the future, this field still needs to address the issues of the reliability and security of the code generated by LLMs, as well as the insufficient explainability of the results of automated performance analysis. The future research focus in this field lies in addressing issues such as the lack of interpretability and insufficient domain knowledge of LLMs. It is necessary to prioritize enhancing the reliability of AI recommendations and promoting the transformation of AI from an auxiliary tool to an intelligent Agent with self-repair capabilities, in order to achieve a truly efficient and secure human-machine collaboration paradigm. This article systematically reviews the relevant progress, aiming to promote the transformation of software engineering from an artificial-driven model to an AI-enhanced automated paradigm. It provides theoretical references for ensuring the quality of backend code, improving product delivery speed, and enhancing system reliability.

Within the pharmaceutical industry, there is an increasing pressure on pharmaceutical companies to submit clinical trial data that fulfill the strict regulatory requirements and operate within tight timeframes and limited resources. The manual generation of CDISC-conformant datasets is still resource-intensive, subject to error, and implicates the generation process as the complexity of a trial increases. Transformative solutions are provided in automation frameworks based on the use of SAS macros and R scripts that are applied to develop metadata-driven development, modular architecture, and dynamic code generation that is dynamic. These frameworks save radically programming time and, at the same time, enhance the data quality metrics, lengths of CDISC conformance, cross-dataset consistency, and specification compliance dimensions. Practical applications show efficiency improvements that allow an organization to handle non-proportional program resource demands. The automation migration needs organizational dedication, tactical planning, and up-front investment in the formation of sound structures, broad metatag designing, and validation mechanisms. Pharmaceutical corporations and educational medical facilities have demonstrated that automation has enabled quicker study completion schedules, lower operational expenses, enhanced regulatory standards, and better contentment of programmers. The Hybrid SAS-R workflows are based on the synergistic use of platform strengths, where regulatory familiarity is provided by SAS, and modern programming capabilities are provided by R. Techniques of performance optimization, such as parallel processing, incremental updates as well and effective data structures make ensure that the frameworks can be scaled easily with the increase of the data volumes. Among the success factors, one can identify the initiation of focused pilot implementations, investment in metadata quality, emphasis on validation, documentation, creation of cross-functional collaboration, and formal governance. Automation will enable statistical programmers to become strategic consultants and not merely tactical code generators, and enable the intellectual power to develop novel analytical techniques and strategic advice to clinical teams that assist in generating the evidence needed to make regulatory decisions.

This article discusses the importance of effective communication tools in public health, highlighting innovations like Quick Response (QR) codes and QR wallet reference cards (QR cards) for enhancing outreach and education. QR codes are scannable barcodes that link to digital content. QR cards are compact cards, similar to business cards, with codes that lead to relevant health information. To our knowledge, there is little published literature on using QR codes and cards for public health programs and health communication outside of health care clinics and education settings. The North Carolina Department of Health and Human Services, Division of Public Health, Community and Clinical Connections for Prevention and Health Branch has successfully implemented QR codes in various public health programs, particularly in diabetes management and nutrition, physical activity, and obesity initiatives. Key lessons learned include using reputable QR code generators, ensuring visibility and scanability of the codes, testing links before use, providing clear calls to action, and considering dynamic versus static codes based on needs. QR codes can be leveraged in public health practice for program promotion, evaluation sharing, and community resource accessibility. However, limitations such as smartphone dependency among some populations should be acknowledged. In conclusion, while QR codes are a simple tool, they hold significant potential for improving public health communication. Research on QR code use in public health settings could help inform best practices for public health programs and health promotion across different contexts.

Large Language Models (LLMs) demonstrate potential in code generation capabilities, yet their applicability in autonomous vehicle control has not been sufficiently explored. This study verifies whether LLMs can generate executable MATLAB code for software-defined vehicle scenarios, comparing five models: GPT-4, Gemini 2.5 Pro, Claude Sonnet 4.0, CodeLlama-13B-Instruct, and StarCoder2. Thirteen standardised prompts were applied across three types of scenarios: programming-based driving scenarios, inertial sensor-based simulations, and vehicle parking scenarios. Multiple automated evaluation metrics—BLEU, ROUGE-L, ChrF, Spec-Compliance, and Runtime-Sanity—were used to assess code executability, accuracy, and completeness. The results showed GPT-4 achieved the highest score 0.54 in the parking scenario with an overall average score of 0.27, followed by Gemini 2.5 Pro as 0.26. Commercial models demonstrated over 60% execution success rates across all scenarios, whereas open-source models like CodeLlama and StarCoder2 were limited to under 20%. Furthermore, the parking scenario yielded the lowest average score of 0.19, confirming that complex tasks involving sensor synchronisation and trajectory control represent a common limitation across all models. This study presents a new benchmark for quantitatively evaluating the quality of SDV control code generated by LLMs, empirically demonstrating that prompt design and task complexity critically influence model reliability and real-world applicability.

ABSTRACT This study explores matrix‐free tangent evaluations in finite‐strain elasticity with the use of automatically generated code for the quadrature‐point level calculations. The code generation is done via automatic differentiation (AD) with AceGen. We compare hand‐written and AD‐generated codes under two computing strategies: on‐the‐fly evaluation and caching intermediate results. The comparison reveals that the AD‐generated code achieves superior performance in matrix‐free computations.

AI4SWEng, a Horizon Europe project that will be active from 2025 until 2028, unites 15 leading partners across the European Union, Switzerland and Turkey, combining experts in Model-Driven Software Engineering and trustworthy AI, with a special focus on applying Large Language Models. The project addresses complex challenges in industries such as healthcare, cyber-physical systems, and electric vehicles, focusing on multi-architectural and resource-constrained systems. Our mission is to transform agile software development by leveraging AI to boost efficiency, reliability, and security while ensuring ethical and regulatory compliance. The goal is to deliver scalable, sustainable, and socially responsible solutions that accelerate time-to-market without compromising quality. The AI4SWEng project will deliver an AI-powered software engineering suite providing end-to-end support for the software lifecycle, from code generation and advanced debugging to security, energy efficiency, and project management. By reducing pain points and enhancing productivity, the suite/platform aims to reduce developer stress, foster creativity, and improve job satisfaction. With a commitment to user-centred design and advanced prompt engineering, we empower developers to harness the full potential of AI. In this paper the AI4SWEng project consortium presents a novel and strategic approach proposed by the project consortium that will shape the future of software engineering, driving innovation and paving the way for more agile, intelligent, and sustainable software development.

Hiperti: high performance system for cross-platform code generation of transformer model inference based on MLIR

Abstract Conversational large-language models (LLMs), such as ChatGPT, are extensively used for issue resolution tasks, particularly for generating ideas to implement new features or resolve bugs. However, not all developer-LLM conversations are useful for effective issue resolution and it is still unknown what makes some of these conversations not helpful. In this paper, we analyze 686 developer-ChatGPT conversations shared within GitHub issue threads to identify characteristics that make these conversations effective for issue resolution. First, we empirically analyze the conversations and their corresponding issue threads to distinguish helpful from unhelpful conversations. We begin by categorizing the types of tasks developers seek help with (e.g., code generation , bug identification and fixing , test generation ), to better understand the scenarios in which ChatGPT is most effective. Next, we examine a wide range of conversational, project, and issue-related metrics to uncover statistically significant factors associated with helpful conversations. Finally, we identify common deficiencies in unhelpful ChatGPT responses to highlight areas that could inform the design of more effective developer-facing tools. We found that only 62% of the ChatGPT conversations were helpful for successful issue resolution. Among different tasks related to issue resolution, ChatGPT was most helpful in assisting with code generation, and tool/library/API recommendations, but struggled with generating code explanations. Our conversational metrics reveal that helpful conversations are shorter, more readable, and exhibit higher semantic and linguistic alignment. Our project metrics reveal that larger, more popular projects and experienced developers benefit more from ChatGPT’s assistance. Our issue metrics indicate that ChatGPT is more effective on simpler issues characterized by limited developer activity and faster resolution times. These typically involve well-scoped technical problems such as compilation errors and tool feature requests. In contrast, it performs less effectively on complex issues that demand deep project-specific understanding, such as system-level code debugging and refactoring. The most common deficiencies in unhelpful ChatGPT responses include incorrect information and lack of comprehensiveness. Our findings have wide implications including guiding developers on effective interaction strategies for issue resolution, informing the development of tools or frameworks to support optimal prompt design, and providing insights on fine-tuning LLMs for issue resolution tasks.