Frontier LLM agents are automating many business tasks, but their high inference cost makes large-scale deployment unsustainable. Small language models (SLMs) offer a cheaper alternative, yet they typically fall short when swapped into a harness designed for a frontier LLM. We show that for many routine business tasks, SLM agents can match LLM performance at 90

Responsible AI (RAI) tools—checklists, templates, and governance processes—often engage RAI champions, individuals intrinsically motivated to advocate ethical practices, but fail to reach non-champions, who frequently dismiss them as bureaucratic tasks. To explore this gap, we shadowed meetings and interviewed data scientists at an organization, finding that practitioners perceived RAI as irrelevant to their work. Building on these insights and theoretical foundations, we derived design principles for engaging non-champions, and introduced sticky stories—narratives of unexpected ML harms designed to be concrete, severe, surprising, diverse, and relevant, unlike widely circulated media to which practitioners are desensitized. Using a compound AI system, we generated and evaluated sticky stories through human and LLM assessments at scale, confirming they embodied the intended qualities. In a study with 29 practitioners, we found that, compared to regular stories, sticky stories significantly increased the engagement time on harm identification, broadened the range of harms recognized, and fostered deeper reflection.

Coding agents now author entire pull requests, and practitioners sharply disagree about what this does to code review: whether it becomes the bottleneck, whether human review is still necessary, and whether it quietly erodes the understanding that it once built. Repository-mining studies measure surface trends but seldom explain the mechanisms beneath them, and the trends themselves prove unstable. A motivating observational analysis of public GITHUB activity finds that agent-authored pull requests are reviewed less often, merged several times faster, and discussed less than human-authored ones, yet the direction of these trends flips under different but equally defensible analysis choices, so the traces establish what is changing without explaining why. To recover the mechanisms, we synthesize practitioner discourse at scale into an explanatory theory: we collect 38,709 grey-literature documents (engineering blogs and Reddit threads), filter to those substantively about code review, and code a stratified random sample of 3,100 with an LLM-assisted pipeline, from which we build a causal model of 26 constructs and 67 relationships (64 directed, 3 contested). Its organizing claim is that review is the control point through which a coding agent's effect on software is decided, and that AI does not fix the sign of that effect: the team sets it, through the expertise its humans bring and how it structures the review process. The theory makes the competing positions explicit and turns "AI is changing code review" into falsifiable propositions with named constructs and moderators. As a secondary contribution, we offer the underlying LLM-assisted, grey-literature theory-building method as a scalable template for software-engineering research, with a public implementation.

There is increasing interest in integrating AI agents that invoke tools into domain-specific commercial software, where unintended tool calls can cause serious security and safety incidents. This has drawn growing research attention, and many agent security and safety benchmarks have emerged. They implicitly shape how the community approaches security and safety. Yet existing work exhibits a blind spot: it emphasizes training-based methods and neural guardrails, which reduce the likelihood of insecure or unsafe actions but cannot guarantee their prevention. It generally overlooks opportunities for deductive, symbolic guardrails grounded in standard software engineering practices, which can provide guarantees for some security and safety requirements. Our study has three parts: (1) a systematic review of 80 agent security and safety benchmarks finding that that 85% of benchmarks do not state verifiable requirements (61% provide none, and 24% give only high-level goals); (2) an applicability analysis of which security and safety requirements symbolic guardrails can and cannot enforce on τ^2-Bench, CAR-bench, and MedAgentBench, finding that 74% of requirements are symbolically enforceable and 95% of these need only simple, low-cost checks; and (3) an empirical evaluation of symbolic guardrails on the same three benchmarks, finding that symbolic guardrails improve security and safety without sacrificing utility, and often improve it. Our work draws attention to the potential for symbolic guardrails for AI agents, suggesting them as an overlooked but practical path toward deploying domain-specific AI agents in risk-averse commercial software. We release all codes and artifacts at https://github.com/hyn0027/agent-symbolic-guardrails.

Peer review is essential to software engineering research, yet reviewer training remains largely implicit. We describe the ICSE 2026 Shadow PC, a redesigned program emphasizing deliberate practice at scale. Key innovations include multi-phase structure with calibration and peer feedback, strict separation from the main PC, and a pathway toward leadership development. With 102 participants completing the program and reviewing 117 papers, 97

Prompt underspecification is a common challenge when interacting with LLMs. In this paper, we present an in-depth analysis of this problem, showing that while LLMs can often infer unspecified requirements by default (41.1

Large language models (LLMs) have demonstrated the promise to revolutionize the field of software engineering. Among other things, LLM agents are rapidly gaining momentum in software development, with practitioners reporting a multifold increase in productivity after adoption. Yet, empirical evidence is lacking around these claims. In this paper, we estimate the causal effect of adopting a widely popular LLM agent assistant, namely Cursor, on development velocity and software quality. The estimation is enabled by a state-of-the-art difference-in-differences design comparing Cursor-adopting GitHub projects with a matched control group of similar GitHub projects that do not use Cursor. We find that the adoption of Cursor leads to a statistically significant, large, but transient increase in project-level development velocity, along with a substantial and persistent increase in static analysis warnings and code complexity. Further panel generalized-method-of-moments estimation reveals that increases in static analysis warnings and code complexity are major factors driving long-term velocity slowdown. Our study identifies quality assurance as a major bottleneck for early Cursor adopters and calls for it to be a first-class citizen in the design of agentic AI coding tools and AI-driven workflows.

Machine learning (ML) components are increasingly incorporated into software products for end-users, but developers face challenges in transitioning from ML prototypes to products. Academics have limited access to the source of commercial ML products, hindering research progress to address these challenges. In this study, first and foremost, we contribute a dataset of 262 open-source ML products for end users (not just models), identified among more than half a million ML-related projects on GitHub. Then, we qualitatively and quantitatively analyze 30 open-source ML products to answer six broad research questions about development practices and system architecture. We find that the majority of the ML products in our sample represent more startup-style development than reported in past interview studies. We report 21 findings, including limited involvement of data scientists in many open-source ML products, unusually low modularity between ML and non-ML code, diverse architectural choices on incorporating models into products, and limited prevalence of industry best practices such as model testing, pipeline automation, and monitoring. Additionally, we discuss seven implications of this study on research, development, and education, including the need for tools to assist teams without data scientists, education opportunities, and open-source-specific research for privacy-preserving telemetry.

Highly-configurable software often includes performance-sensitive configuration options. There are performance expectations in different configurations, but these expectations may not hold due to inaccurate mental models, corner cases, or unanticipated interactions with other options. We propose differential performance fuzzing of configuration options, a fuzzing technique that uses differential performance feedback to automatically identify inputs that violate these expectations for specific configuration changes. By guiding fuzzing toward scenarios where a supposedly faster configuration performs worse, differential performance fuzzing reveals unexpected performance behavior effectively. In our preliminary evaluation, our method identified unexpected performance gains in configurations presumed slower for 4 configuration options in Closure, demonstrating the potential for detecting performance issues in real-world applications.

Large Language Models (LLMs) are increasingly embedded into software products across diverse industries, enhancing user experiences, but at the same time introducing numerous challenges for developers. Unique characteristics of LLMs force developers, who are accustomed to traditional software development and evaluation, out of their comfort zones as the LLM components shatter standard assumptions about software systems. This study explores the emerging solutions that software developers are adopting to navigate the encountered challenges. Leveraging a mixed-method research, including 26 interviews and a survey with 182 responses, the study identifies 19 emerging solutions regarding quality assurance that practitioners across several product teams at Microsoft are exploring. The findings provide valuable insights that can guide the development and evaluation of LLM-based products more broadly in the face of these challenges.

Performance-influence models are beneficial for understanding how configurations affect system performance, but their creation is challenging due to the exponential growth of configuration spaces. While gray-box approaches leverage selective "structural knowledge" (like the module execution graph of the system) to improve modeling, the relationship between this knowledge, a system's characteristics (we call them "structural aspects"), and potential model improvements is not well understood. This paper addresses this gap by formally investigating how variations in structural aspects (e.g., the number of modules and options per module) and the level of structural knowledge impact the creation of "opportunities" for improved "modular performance modeling". We introduce and quantify the concept of modeling "hardness", defined as the inherent difficulty of performance modeling. Through controlled experiments with synthetic system models, we establish an "analytical matrix" to measure these concepts. Our findings show that modeling hardness is primarily driven by the number of modules and configuration options per module. More importantly, we demonstrate that both higher levels of structural knowledge and increased modeling hardness significantly enhance the opportunity for improvement. The impact of these factors varies by performance metric; for ranking accuracy (e.g., in debugging task), structural knowledge is more dominant, while for prediction accuracy (e.g., in resource management task), hardness plays a stronger role. These results provide actionable insights for system designers, guiding them to strategically allocate time and select appropriate modeling approaches based on a system's characteristics and a given task's objectives.

Developers play a central role in determining how machine learning systems are explained in practice, yet they are rarely trained to design explanations for non-technical audiences. Despite this, transparency and explainability requirements are increasingly codified in regulation and organizational policy. It remains unclear how such policies influence developer behavior or the quality of the explanations they produce. We report results from two controlled experiments with 194 participants, typical developers without specialized training in human-centered explainable AI, who designed explanations for an ML-powered diabetic retinopathy screening tool. In the first experiment, differences in policy purpose and level of detail had little effect: policy guidance was often ignored and explanation quality remained low. In the second experiment, stronger enforcement increased formal compliance, but explanations largely remained poorly suited to medical professionals and patients. We further observed that across both experiments, developers repeatedly produced explanations that were technically flawed or difficult to interpret, framed for developers rather than end users, reliant on medical jargon, or insufficiently grounded in the clinical decision context and workflow, with developer-centric framing being the most prevalent. These findings suggest that policy and policy enforcement alone are insufficient to produce meaningful end-user explanations and that responsible AI frameworks may overestimate developers' ability to translate high-level requirements into human-centered designs without additional training, tools, or implementation support.

Machine learning in production needs to balance multiple objectives: This is particularly evident in ranking or recommendation models, where conflicting objectives such as user engagement, satisfaction, diversity, and novelty must be considered at the same time. However, designing multi-objective rankers is inherently a dynamic wicked problem - there is no single optimal solution, and the needs evolve over time. Effective design requires collaboration between cross-functional teams and careful analysis of a wide range of information. In this work, we introduce Orbit, a conceptual framework for Objective-centric Ranker Building and Iteration. The framework places objectives at the center of the design process, to serve as boundary objects for communication and guide practitioners for design and evaluation. We implement Orbit as an interactive system, which enables stakeholders to interact with objective space directly and supports real-time exploration and evaluation of design trade-offs. We evaluate Orbit through a user study involving twelve industry practitioners, showing that it supports efficient design space exploration, leads to more informed decision-making, and enhances awareness of the inherent trade-offs of multiple objectives. Orbit (1) opens up new opportunities of an objective-centric design process for any multi-objective ML models, as well as (2) sheds light on future designs that push practitioners to go beyond a narrow metric-centric or example-centric mindset.

Machine learning (ML) components are increasingly integrated into software products, yet their complexity and inherent uncertainty often lead to unintended and hazardous consequences, both for individuals and society at large. Despite these risks, practitioners seldom adopt proactive approaches to anticipate and mitigate hazards before they occur. Traditional safety engineering approaches, such as Failure Mode and Effects Analysis (FMEA) and System Theoretic Process Analysis (STPA), offer systematic frameworks for early risk identification but are rarely adopted. This position paper advocates for integrating hazard analysis into the development of any ML-powered software product and calls for greater support to make this process accessible to developers. By using large language models (LLMs) to partially automate a modified STPA process with human oversight at critical steps, we expect to address two key challenges: the heavy dependency on highly experienced safety engineering experts, and the time-consuming, labor-intensive nature of traditional hazard analysis, which often impedes its integration into real-world development workflows. We illustrate our approach with a running example, demonstrating that many seemingly unanticipated issues can, in fact, be anticipated.

Reusable software libraries, frameworks, and components, such as those provided by open-source ecosystems and third-party suppliers, accelerate digital innovation. However, recent years have shown almost exponential growth in attackers leveraging these software artifacts to launch software supply chain attacks. Past well-known software supply chain attacks include the SolarWinds, log4j, and xz utils incidents. Supply chain attacks are considered to have three major attack vectors: through vulnerabilities and malware accidentally or intentionally injected into open-source and third-party dependencies/components/containers ; by infiltrating the build infrastructure during the build and deployment processes; and through targeted techniques aimed at the humans involved in software development, such as through social engineering. Plummeting trust in the software supply chain could decelerate digital innovation if the software industry reduces its use of open-source and third-party artifacts to reduce risks. This paper contains perspectives and knowledge obtained from intentional outreach with practitioners to understand their practical challenges and from extensive research efforts. We then provide an overview of current research efforts to secure the software supply chain. Finally, we propose a future research agenda to close software supply chain attack vectors and support the software industry.

Algorithmic fairness of machine learning (ML) models has raised significant concern in the recent years. Many testing, verification, and bias mitigation techniques have been proposed to identify and reduce fairness issues in ML models. The existing methods are model-centric and designed to detect fairness issues under static settings. However, many ML-enabled systems operate in a dynamic environment where the predictive decisions made by the system impact the environment, which in turn affects future decision-making. Such a self-reinforcing feedback loop can cause fairness violations in the long term, even if the immediate outcomes are fair. In this paper, we propose a simulation-based framework called FairSense to detect and analyze long-term unfairness in ML-enabled systems. Given a fairness requirement, FairSense performs Monte-Carlo simulation to enumerate evolution traces for each system configuration. Then, FairSense performs sensitivity analysis on the space of possible configurations to understand the impact of design options and environmental factors on the long-term fairness of the system. We demonstrate FairSense's potential utility through three real-world case studies: Loan lending, opioids risk scoring, and predictive policing.

One of the major challenges in the use of opaque, complex AI models is the need or desire to provide an explanation to the end-user (and other stakeholders) as to how the system arrived at the answer it did. While there is significant research in the development of explainability techniques for AI, the question remains as to who needs an explanation, what an explanation consists of, and how to communicate this to a lay user who lacks direct expertise in the area. In this position paper, an interdisciplinary team of researchers argue that the example of clinical communications offers lessons to those interested in improving the transparency and interpretability of AI systems. We identify five lessons from clinical communications: (1) offering explanations for AI systems and disclosure of their use recognizes the dignity of those using and impacted by it; (2) AI explanations can be productively targeted rather than totally comprehensive; (3) AI explanations can be enforced through codified rules but also norms, guided by core values; (4) what constitutes a “good” AI explanation will require repeated updating due to changes in technology and social expectations; 5) AI explanations will have impacts beyond defining any one AI system, shaping and being shaped by broader perceptions of AI. We review the history, debates and consequences surrounding the institutionalization of one type of clinical communication, informed consent, in order to illustrate the challenges and opportunities that may await attempts to offer explanations of opaque AI models. We highlight takeaways and implications for computer scientists and policymakers in the context of growing concerns and moves toward AI governance.

Supply chain security has become a very important vector to consider when defending against adversary attacks. Due to this, more and more developers are keen on improving their supply chains to make them more robust against future threats. On March 7th, 2024 researchers from the Secure Software Supply Chain Center (S3C2) gathered 14 industry leaders, developers and consumers of the open source ecosystem to discuss the state of supply chain security. The goal of the summit is to share insights between companies and developers alike to foster new collaborations and ideas moving forward. Through this meeting, participants were questions on best practices and thoughts how to improve things for the future. In this paper we summarize the responses and discussions of the summit. The panel questions can be found in the appendix.

With the rise of artificial intelligence (AI), concerns about AI applications causing unforeseen harms to safety, privacy, security, and fairness are intensifying. While attempts to create regulations are underway, with initiatives such as the EU AI Act and the 2023 White House executive order, skepticism abounds as to the efficacy of such regulations. This paper explores an interdisciplinary approach to designing policy for the explainability of AI applications, as the widely discussed "right to explanation" associated with the EU General Data Protection Regulation is ambiguous. To develop practical guidance for explainability, we conducted an experimental study that involved continuous collaboration among a team of researchers with AI and policy backgrounds over the course of ten weeks. The objective was to determine whether, through interdisciplinary effort, we can reach consensus on a policy for explainability in AI–one that is clearer, and more actionable and enforceable than current guidelines. We share nine observations, derived from an iterative policy design process, which included drafting the policy, attempting to comply with it (or circumvent it), and collectively evaluating its effectiveness on a weekly basis. Key observations include: iterative and continuous feedback was useful to improve policy drafts over time, discussing evidence of compliance was necessary during policy design, and human-subject studies were found to be an important form of evidence. We conclude with a note of optimism, arguing that meaningful policies can be achieved within a moderate time frame and with limited experience in policy design, as demonstrated by our student researchers on the team. This holds promising implications for policymakers, signaling that practical and effective regulation for AI applications is attainable.