AI-accelerated code widens gap between software design & reality

Research from vFunction has found that discrepancies between software architecture documentation and production reality present significant business risks, especially as AI-accelerated code creation becomes more prevalent in enterprise development.

The study, titled "2025 Architecture in Software Development," surveyed more than 600 IT professionals and identified a notable disconnect between executive perceptions of architecture integration and the actual state of software production. While 63% of organisations claim that architecture is integrated throughout their software development life cycle, over half (56%) acknowledge that their production environments do not match the documented architecture.

This architectural misalignment has led to serious business consequences, with 53% of respondents reporting project delays, 50% encountering security or compliance challenges, and 32% experiencing service disruptions. A further 47% noted that these inconsistencies have resulted in unexpected operational costs.

Moti Rafalin, CEO and co-founder of vFunction, commented on the broader implications for businesses: "When architectural documentation diverges from reality, businesses suffer tangible consequences. Our research shows this isn't just a technical problem. In fact, 47% of organisations reported unexpected operational costs tied to misalignment between their documented architecture and what's actually implemented. This kind of disconnect directly impacts efficiency, security, scalability, and ultimately, the bottom line."

The study also highlights a perception gap between different roles within organisations. While 52% of executives believe that their documentation is fully aligned, only 36% of practitioners share this view. Executive recognition of business impacts is higher, with 70% acknowledging project delays due to misalignment, as opposed to 49% of practitioners, who focus more on technical challenges. In industries such as financial services and manufacturing, alignment is even lower - only 31% and 34% respectively describe their documentation as fully matching production.

Organisational size also affects alignment, with smaller companies (with revenue of $100 million to $999 million) reportedly having better documentation quality (52% fully aligned) compared to larger enterprises (over $1 billion), where only 40% noted full alignment.

The impact of this misalignment is particularly pronounced in highly regulated sectors, such as financial services, in which 50% of respondents cited security and compliance issues as being the top misalignment concern. More broadly, 93% of respondents indicated experiencing negative outcomes, including disruptions to service, high operational costs, and security challenges, linked to gaps between architecture and documentation.

Teams surveyed placed a strong emphasis on the need for better documentation tools (69%), enhanced architectural governance (69%), and improved architecture integration throughout the development process (66%). Executives prioritise governance and documentation, while practitioners emphasise collaboration between architects and development teams to bridge this divide.

Rafalin commented on these priorities: "Successful architecture integration across the SDLC requires merging the perspectives of both leaders and practitioners - combining top-down vision with frontline expertise. When leadership pushes governance and tools while practitioners champion collaboration, both sides miss the full picture. Aligning these priorities ensures that architecture initiatives succeed, delivering value across the organisation."

The influence of AI in software development is also driving optimism for simplification, with 65% of respondents believing that AI-accelerated approaches will simplify their current application architecture. 

However, Rafalin warned of the potential risks: "As organisations aggressively adopt AI to automate processes and generate code, they're introducing new layers of complexity into their architecture. AI currently lacks the system-wide view which could lead to code duplication and microservices sprawl, escalating risks in security, scalability, and compliance. Effective governance and continuous observability are essential for controlling the consequences of AI-generated code complexity, enforcing clear architectural boundaries and preventing system failures."

Overall, 90% of respondents see value in integrating architecture insights into observability platforms, with OpenTelemetry being widely adopted - 59% reported using it either exclusively or in combination with other systems.

Rafalin summed up the situation: "The strategic importance of architecture is clear, but without visibility, integration, and continuous management, architecture cannot support business growth. Businesses should be focused on improving observability, using technologies like OpenTelemetry and AI to streamline architecture management and cut through complexity. For architecture to truly serve as a lever for growth and security guardrails, organisations must embrace real-time insights and intelligent tools that make architectural complexity manageable and actionable in daily operations."

Alongside the survey findings, vFunction announced new advancements to its architectural observability platform aimed at helping organisations address these issues. The platform now offers subsystem-level microservices management, component diagrams for monolithic architectures, and integrations with enterprise architecture tools for real-time documentation updates. These new features are designed to provide deeper visibility into distributed architecture and to bridge the gap between rapid development and architectural integrity.

Rafalin stated, "Today's software teams struggle to maintain architectural integrity while delivering new features and functionality at unprecedented speeds, especially as AI-powered development accelerates code creation. Our enhanced platform supports the continuous architecture lifecycle - from design and baseline creation through measurement, validation, and drift remediation - ensuring systems evolve intentionally. By providing comprehensive visibility at different levels of granularity, we're empowering software architects and engineering teams to make informed architectural decisions that balance velocity with quality. This helps companies build systems that can scale, evolve, and support business innovation despite increasing complexity."

Yogi Byreddy, CTO of CTIS, commented on their usage: "As our application environment expands, we're focused on proactively preventing architectural challenges rather than reacting to them. vFunction shows us our architecture components and their interactions in a graphical representation that is easy to comprehend. This visibility has enhanced our approach to modernisation and new development initiatives. Our teams can now collaborate with shared architectural facts, making better decisions based on what's actually happening in our systems."

Amir Rapson, CTO, CCSO and co-founder at vFunction, explained how the platform addresses the risks of rapid AI-assisted development: "While AI coding tools excel at generating functional solutions quickly, they don't inherently understand architectural boundaries or system design principles. Our platform bridges this gap by providing the visibility and governance necessary to ensure that faster development doesn't lead to architectural chaos that hampers future innovation."

James Governor, analyst and co-founder at RedMonk, also highlighted the strategic shift: "With the rise of AI-assisted code generation and the need to increase development velocity, it is increasingly difficult for enterprises to maintain a coherent application and systems architecture model. vFunction is integrating Open Telemetry data into its stack in order to provide real time, constantly updated architectural observability and documentation to reduce sprawl and enable consistent management."

Rapson discussed the company's support for modernising monolithic applications: "Modernising monoliths requires a fundamentally different approach than managing distributed systems. Our teams have worked with some of the world's most complex applications, and we've embedded that expertise directly into our platform to guide organisations through what has traditionally been an overwhelming process."