Generative AI is revolutionizing software development, offering faster coding, automation, and efficiency. However, experts at a recent IIIT Hyderabad roundtable stressed that while AI accelerates the process, it still requires significant human oversight. The discussion, which included researchers, technologists, and industry leaders from banking, healthcare, SaaS, and startups, explored both AI’s advantages and limitations in coding.

One of AI’s biggest strengths is its ability to reduce search time and help developers complete tasks more quickly. Tools powered by AI can generate boilerplate code, suggest optimizations, and automate debugging. A trial by the open-source community Swecha found that AI-assisted teams completed tasks nearly twice as fast as those using traditional coding methods. However, AI struggles with complex business logic, legacy systems, and industry-specific applications, with experts noting that AI-generated responses in business logic are incorrect 50–60% of the time, making human intervention necessary.

Another major issue is technical debt—while AI can produce functional code, it often lacks optimization and maintainability. Poorly optimized AI-generated code can introduce inefficiencies, leading to long-term issues. Experts were divided on whether technical debt could be completely eliminated or if managing it was an inevitable challenge of AI-driven development.

Despite the buzz around AI, its adoption among developers remains low, at around 20%, even in companies investing heavily in AI. Many developers remain skeptical, preferring to write code manually, while others worry about job security. Data privacy concerns also hinder adoption, as businesses hesitate to share proprietary information with AI tools due to security and regulatory risks.

Organizations that have successfully integrated AI have done so cautiously, treating it as a junior developer rather than an autonomous problem-solver. Some banking firms reported up to 85% accuracy in AI-generated SQL queries when reviewed by human experts. AI is most effective when used as a supervised assistant rather than a fully independent coder.

Training developers in effective AI use is crucial, as many struggle with writing precise prompts, leading to unreliable outputs. Some companies are conducting workshops and pilot projects to improve AI adoption while developing governance frameworks to regulate its use in security, intellectual property, and accountability.

AI is transforming software development, not by reducing engineering teams, but by shifting focus from routine coding to higher-level problem-solving. New roles, such as AI-driven product managers, are likely to emerge. The key takeaway from the discussion was clear: AI is a tool, not a shortcut. Businesses that trust but verify AI-generated code will benefit the most, ensuring a balance between AI’s speed and the accuracy that only human oversight can provide.

 Key Insights from the Discussion:

• Efficiency vs. Accuracy:
  AI tools accelerate coding but struggle with legacy systems, complex logic, and business-specific applications. Responses are incorrect 50-60% of the time in business logic scenarios.
• Technical Debt and Adoption Challenges:
  AI-generated code is often functional but not optimized, leading to long-term inefficiencies. Adoption rates among developers remain low (~20%), as many prefer to code manually or fear job displacement.
• Security and Privacy Concerns:
  Companies hesitate to share proprietary data with AI models due to regulatory and security risks. AI is used cautiously, with organizations treating it as a junior developer requiring supervision.
• Successful AI Integration:
  Some banking firms reported 85% accuracy in AI-generated SQL queries with human review. Open-source trials showed AI-assisted teams completed tasks twice as fast as traditional methods.
• Training and Governance:
  Developers unfamiliar with AI struggle with precise prompts, leading to unreliable outputs. Companies are establishing governance frameworks to regulate AI use in security, IP, and accountability.

Conclusion:

Experts emphasized that AI is a tool, not a shortcut. It reshapes development roles, shifting focus from routine coding to higher-level problem-solving. While AI won’t replace human programmers, new roles such as AI-driven product managers may emerge to oversee AI-generated code.