AI integration in software raises architectural challenges

Software development teams are facing a significant shift in engineering practices as artificial intelligence (AI) becomes increasingly integrated into the development process.

Moti Rafalin, the Chief Executive Officer and Co-founder of vFunction, has highlighted how AI advancements will accelerate code writing and software creation, but simultaneously reveal weaknesses in existing architectures. He suggested this could lead to a proliferation of microservices and heightened complexity within systems.

Rafalin stated, "While AI can make code development simpler and faster, it won't solve architectural challenges. When it comes to building, improving, and fixing applications, AI tools like ChatGPT and other large language models cannot effectively address issues in the interactions between components."

He further explained that these AI tools are proficient in generating individual code components, yet they struggle to comprehend the broader system interactions, thereby neglecting scalability and reliability issues, which are more deeply rooted in architectural design than in code quality. "As AI speeds up coding, teams must focus more on system-wide architecture design and documentation," said Rafalin.

The use of AI in creating microservices has been identified as a potential source of complexity due to rapid code generation, which may bypass checks for overlapping functionalities within a system. Rafalin elaborated, "The speed of AI code generation will make it significantly easier to create new microservices, potentially leading to microservices sprawl and increased complexity." He advised that development teams should adopt architecture governance to manage this complexity.

OpenTelemetry, an industry-wide initiative, is seen as an opportunity for organisations to enhance value from their observability solutions. With OpenTelemetry, companies can switch between observability platforms more effortlessly and layer additional functionalities onto their existing systems. "Organizations will extract more value from their observability investments through OpenTelemetry's extensibility," Rafalin commented.

He noted that the focus of observability solutions should also evolve to include architectural insights, facilitating better understanding and management of the architecture-related aspects that impact system performance. "As applications grow more complex, traditional approaches to observability must expand to address architectural challenges," Rafalin remarked.

Amir Rapson, vFunction's Chief Technology Officer and Co-founder, pointed to the inefficacy of the "slow and steady" approach in cloud modernization, advocating for strategies that highlight speed and efficiency. "Organizations will need to break free from gradual modernization patterns and embrace more decisive approaches," he noted.

Rapson additionally sees a shift away from superficial metrics like code quality scores and towards deeper governance that addresses system evolution and mitigates unnecessary complexity. He argued, "In 2025, organizations will finally start to realize that generating perfect code through AI doesn't guarantee good software."

Ori Saporta, Vice President of Engineering at vFunction, expressed concerns about the challenges arising from mass AI-generated code, which could lead to system gridlock due to complex interdependencies among thousands of components. "While AI makes writing code faster, engineering teams will be challenged in 2025 and beyond to take control of their software architecture as thousands of AI-generated components interact," stated Saporta.

Saporta stressed the importance of understanding and optimizing software architecture to prevent a decline in system reliability and to manage operational costs effectively. "Far too many organizations run bloated, complex Frankenstein systems they barely understand and can no longer sustain," he commented.