TLDR: A new study surveyed 55 software practitioners to understand AI adoption in Requirements Engineering (RE). It found that 58.2% of respondents use AI in RE, with 69.1% viewing its impact positively. Human-AI Collaboration (HAIC) is the dominant approach (54.4%), while full AI automation remains minimal. Practitioners value AI as a collaborative partner, but challenges include AI’s lack of domain understanding, communication limitations, and quality concerns. Opportunities lie in AI’s ability to enhance efficiency, intelligent analysis, knowledge augmentation, quality assurance, and documentation. The study highlights the need for RE-specific HAIC frameworks and robust AI governance to unlock AI’s full potential.
Artificial intelligence (AI) is rapidly transforming various industries, and software engineering is no exception. A recent study delves into how AI is being adopted in Requirements Engineering (RE), a crucial phase in software development, and explores the perspectives of practitioners on its use.
Requirements Engineering is the foundation of successful software projects, encompassing four key phases: elicitation (gathering requirements), analysis (understanding and refining them), specification (documenting them), and validation (ensuring they meet needs). With the rise of large language models (LLMs) and AI-powered tools, new opportunities have emerged for these tasks, particularly in processing natural language and automating repetitive work.
The study, titled AI for Requirements Engineering: Industry Adoption and Practitioner Perspectives, surveyed 55 software practitioners to map AI usage across these four RE phases and four approaches to decision-making: human-only, AI validation, Human–AI Collaboration (HAIC), and full AI automation. The researchers also gathered insights into the challenges and opportunities practitioners face when integrating AI into their daily RE work.
Widespread Adoption and Positive Outlook
The findings reveal a significant trend: 58.2% of respondents are already using AI in RE, and a substantial 69.1% view its impact as positive or very positive. This indicates that AI has moved beyond experimental use and is becoming a mainstream tool in the RE process.
Interestingly, the study found that Human–AI Collaboration (HAIC) is the dominant approach, accounting for 54.4% of all RE techniques. This means practitioners largely prefer AI as an active partner rather than a passive validator or a complete replacement for human expertise. Full AI automation remains minimal at 5.4%, suggesting a cautious approach to fully autonomous AI in high-stakes requirements decisions.
The analysis phase showed the highest rate of AI collaboration (60.5%), benefiting most from AI’s analytical capabilities. Elicitation, which involves significant human interaction and emotional intelligence, still maintains higher human decision-making (39.2%).
Responsible AI Practices: A Gap in Governance
While AI adoption is growing, the study highlighted a gap in responsible AI practices. Most practitioners (81.2%) prioritize immediate oversight, such as having humans review and approve AI suggestions. However, systematic governance practices like assessing AI risk levels (37.5%) or keeping records of AI-generated requirements (37.5%) showed lower adoption rates. This suggests a reactive rather than proactive approach to AI governance, potentially leading to long-term accountability and trust issues.
Challenges and Opportunities
Practitioners identified several challenges in using AI for RE:
- Knowledge and Cognitive Limitations: AI models often lack deep domain understanding, contextual awareness, and the ability to grasp unstated or implicit requirements, leading to generic solutions.
- Communication and Human Interaction: AI cannot replicate essential interpersonal skills, build rapport, interpret non-verbal cues, or navigate politically sensitive requirements.
- Quality and Accuracy Concerns: Issues with AI generating inaccurate, incomplete, or generic outputs that compromise project quality.
- Technical and Implementation Barriers: Struggles with complex business logic, integration with existing processes, managing bias, scalability, and token limits in LLMs.
- Data and Methodological Challenges: Limitations in training data specific to RE, and difficulties in tailoring specifications to unique client needs.
- Governance, Security, and Compliance: Concerns about privacy, security, legal risks, explainability, and difficulty in validating against regulatory compliance.
Despite these challenges, significant opportunities were also recognized:
- Efficiency and Automation: AI can automate routine tasks like drafting requirements, summarizing inputs, and identifying gaps, boosting productivity.
- Intelligent Analysis and Pattern Recognition: AI excels at processing large datasets, identifying patterns, suggesting missing requirements, and detecting inconsistencies.
- Knowledge and Expertise Augmentation: AI can act as an intellectual multiplier, combining requirements in novel ways, providing critical reviews, and processing user feedback for new features.
- Quality and Compliance Assurance: AI can improve quality control by cross-referencing requirements against regulations, flagging conflicts, and generating test cases.
- Documentation and Organization: AI can streamline documentation by auto-generating technical specifications and categorizing requirements, ensuring consistency and efficiency.
Also Read:
- Navigating the AI Frontier: A Vision for Generative AI in Software Engineering
- Agentic AI’s True Measure: Why Validation Outweighs Foundation Models
The Future of AI in Requirements Engineering
The study concludes that AI in RE is most effective as a collaborator guided by human expertise. This HAIC approach aligns with theories suggesting that intermediate automation, which combines human oversight with AI capabilities, often yields the best results. The limited adoption of fully autonomous AI underscores the critical role of human judgment in complex, high-stakes RE decisions.
For practitioners, the implication is to focus AI integration on analytical tasks while maintaining human oversight for communication-heavy activities. Researchers are encouraged to investigate barriers to adoption and develop frameworks for optimizing HAIC. AI tool developers should prioritize features that enhance collaboration and augment human capabilities, focusing on trust-building mechanisms, transparency, and domain-specific understanding.
Ultimately, the success of AI in Requirements Engineering hinges on addressing practical barriers, developing robust governance frameworks, and creating AI tools that are specifically designed to understand and support the nuanced contexts of RE.
