As artificial intelligence reshapes the technology sector, Sync NI met with Dr. Vahid Garousi, Senior Lecturer of Software Engineering at Queens University Belfast to discuss how AI is impacting software engineering.
The debate over artificial intelligence's role in software engineering has moved beyond theoretical discussion into practical reality, as engineers worldwide grapple with both the promises and pitfalls of AI-powered tools. While the technology offers unprecedented productivity gains, Vahid warns that its implementation requires careful consideration and human oversight.
Software engineering, a discipline that emerged from a 1968 NATO conference in Germany, has always been about systematic approaches to development, testing and operation of software systems. The term itself was coined by military alliance personnel who reasoned that if mechanical and electrical systems warranted the "engineering" designation, so too should software. Today, that systematic discipline faces perhaps its most significant transformation since its inception.
READ MORE: How Rapid7 and Queen's University Are Forging The Future For Cyber Security
The intersection of AI and software engineering presents a nuanced picture, far removed from simplistic narratives of technological salvation or doom. According to Vahid, AI's impact depends critically on context, implementation, and the quality of human guidance of AI.
The promise of AI-enhanced productivity has captivated the software industry, yet the reality proves more complex than initial enthusiasm suggested. When deployed appropriately, AI tools have the potential to significantly boost efficiency across the software development lifecycle, from requirements gathering through to testing and deployment. However, misapplication of these same tools can lead to disappointing results and, in some cases, actually slow development rather than accelerate it.
As Vahid explains, the quality of AI-generated output emerges as a crucial factor. When provided with proper context and clear instructions by humans operating them, AI systems can produce high-quality code and test cases that genuinely advance projects. Conversely, poorly specified requests would yield substandard outputs that demand extensive correction, potentially negating any time savings.
This challenge connects to a fundamental limitation of current AI systems: their tendency toward what is widely known as "hallucination." Unlike humans who typically acknowledge gaps in their knowledge, AI tools often generate plausible-sounding but fundamentally flawed outputs. This phenomenon has manifested across sectors, with reports emerging of legal professionals discovering that AI-generated case law citations referred to non-existent cases.
The verification problem this creates presents a troubling paradox. Thoroughly checking AI outputs can consume as much time as completing tasks manually from the outset, undermining the efficiency gains that justified adopting the technology initially.
Vahid outlines several examples where AI excels, and also where the current technology had fundamental shortcomings.
Pattern recognition and repetitive tasks represent AI's current advantages in software engineering. The technology can provide tangible productivity gains in areas where it has been
extensively trained and where tasks follow established patterns. These capabilities prove particularly valuable in certain phases of the software development lifecycle.
Vahid provides multiple examples where human capabilities are still shown to be irreplaceable by AI. Creativity in various in software engineering tasks such as user interface (UI) design, feature brainstorming, and user experience (UX) optimisation still require distinctly human judgment. The nuanced decisions that make software genuinely user-friendly rather than merely functional continue to elude AI systems, for now.
Ethical judgment represents another critical gap. The classic thought experiment of autonomous vehicles facing impossible choices such as determining which pedestrian to strike when collision becomes unavoidable illustrates the profound moral dimensions that current AI cannot navigate. Similar dilemmas arise in software designed for healthcare, where the manner of delivering sensitive information or providing a life changing diagnosis requires empathy and cultural awareness.
These limitations have sparked serious discussions around responsible and ethical AI development. As artificial intelligence extends into military applications, with self-flying and self-targeting aircraft already operational, the importance of human oversight intensifies.
To illustrate the point Vahid referenced the autonomous vehicle industry's five-level classification system, ranging from simple driver assistance to complete autonomy. This provides a useful framework for understanding AI's evolving role in software development. Currently, most implementations function at lower autonomy levels, with AI suggesting approaches that human developers can accept or reject.
This recommendation-system approach preserves human agency whilst providing AI-generated insights. Developers review suggestions as they might consider advice from a colleague, incorporating useful ideas whilst dismissing inappropriate ones. As AI capabilities mature, some organisations experiment with higher autonomy levels, though full replacement of human developers remains a distant prospect.
The question of job displacement, however, cannot be dismissed entirely. Vahid observed that we read daily in the media how workforce reductions have already begun in some companies, with estimates suggesting that between ten and twenty percent of roles may be affected, depending on the organisation and sector. A recent Channel 4 documentary explored this theme across professions, highlighting widespread uncertainty about AI's employment impact.
As a result, educational institutions have already begun recalibrating their approach to software engineering education. The recognition that AI will handle certain tasks has prompted discussions about which skills students need to develop. That being said, Vahid firmly believes that human oversight of AI systems will remain essential for years, if not decades, creating demand for professionals who can effectively manage and direct artificial intelligence tools.
This oversight role extends beyond simply checking AI outputs. It encompasses the iterative process of refining AI-generated work, much like directing a junior colleague through successive improvements. The ability to recognise what AI has done well, identify deficiencies, and communicate necessary corrections represents a valuable skill set.
Regulatory frameworks are evolving to formalise this human-in-the-loop approach. AI legislation taking shape across multiple jurisdictions explicitly mandates human oversight,
acknowledging both the technology's limitations and the irreducible need for human judgment in critical decisions.
The impact and utility of AI in software engineering varies considerably across industries. Software serving different sectors including banking, gaming and aviation all face distinct requirements regarding reliability, safety, and regulatory compliance. Consequently, the appropriate level of AI autonomy and the pace of adoption differ markedly between domains.
Life-critical systems, such as those controlling aircraft or medical devices, demand exceptional reliability and extensive testing. The stakes involved in potential failures make human oversight particularly crucial. Conversely, less critical applications might accommodate greater AI autonomy without unacceptable risk.
Local software engineering communities within Northern Ireland's technology sector, such as Software NI and bodies such as the Artificial Intelligence Collaboration Centre (AICC) are actively engaged in discussions about best practices for AI integration. These conversations address practical questions about which tasks benefit from AI assistance and how to maintain quality standards whilst embracing new tools.
Beyond productivity gains in software engineering itself, AI's most significant contributions according to Vahid may lie in addressing humanity's grand challenges. Climate change modelling represents one particularly promising application. Sophisticated AI-powered systems can simulate the impact of policy changes, helping governments navigate the difficult balance between economic growth and environmental protection.
These climate models support strategic planning toward ambitious targets, such as the UK's aim for carbon neutrality. By processing vast quantities of data and modelling countless scenarios, AI can illuminate pathways that might otherwise remain obscure.
According to Vahid, healthcare in particular represents another domain where AI's analytical capabilities could prove transformative. The recent pandemic demonstrated the importance of rapid vaccine development. AI systems working continuously on drug discovery and testing could accelerate responses to future health emergencies, potentially saving countless lives through earlier interventions.
Currently professionals are developing nuanced understandings of when and how to deploy these tools effectively. The technology serves as an aid rather than a panacea, extremely powerful when used appropriately, but requiring skill and judgment to apply successfully.
Vahid believes we will likely see continued evolution in the relationship between human software engineers and AI systems. Autonomy levels may increase in some contexts whilst remaining constrained in others. Educational programmes will adapt to prepare students for a landscape where working alongside AI represents standard practice.
READ MORE: Resilient but stretched: NI Chamber and QUB report highlights softening business conditions
What seems certain is that software engineering will remain a fundamentally human endeavour for the foreseeable future. The discipline's creative aspects, ethical dimensions, and need for empathetic design ensure that human expertise remains at the core. AI will continue to augment and enhance rather than replace software engineers.
For Vahid, AI will not simply replace human developers or leave the profession unchanged. Instead, it introduces a new dynamic, one requiring adaptation, oversight, and thoughtful integration. The challenge facing the industry lies not in resisting change but in shaping it
wisely, ensuring that technological advancement serves humanity whilst preserving the irreplaceable value of human judgment and creativity.
Sync NI's Autumn 2025 magazine explores how Northern Ireland is moving AI from pilot to production
This issue features exclusive insights from industry leaders on real-world AI implementation, responsible adoption, enterprise infrastructure, and how data-driven innovation is transforming businesses across the region.
Read the Big Data Special Edition free online →
Stay connected with NI's tech community:
Subscribe to our newsletter – Get the latest tech news, job opportunities, and events delivered to your inbox
Visit Sync NI – Your daily source for Northern Ireland technology news
