Software developers spend far more time reading and understanding existing code than writing new lines from scratch. Maintenance, debugging, and feature development all depend on navigating what’s already there.
This reality makes readability just as important as productivity. Clear structure, meaningful variable names, and consistent patterns are not just aesthetic choices. They are what allow teams to collaborate and keep systems healthy over time. So with the dawn of AI-powered coding tools, human oversight matters more than ever.
Why code reviews matter more in an AI world
The role of the software developer is shifting. Like it or not, AI has become part of the everyday developer toolkit, enhancing productivity and lowering barriers to development. It’s a powerful change, which introduces new risks. But no matter how the code is produced, ultimate responsibility still rests with people, or as Amazon CTO, Werner Vogels, puts it, “The work is yours.”
Vogels also argues that in this new reality, code reviews are becoming more important than ever before. AI generates and checks code faster, but reviews are not just about catching errors. They’re also essential learning processes within teams. “When code moves faster, shared understanding becomes even more critical,” says Vogels.
Boncode CEO, Jan Willem Klerkx, agrees. In his experience, Vogels’ warning is a present reality. It’s already visible in enterprise software portfolios. Lines of code may increase as a result of AI, but quantity does not equate with quality. “Increased output makes developer discipline more important than ever before, especially for teams adjusting to managing AI-assisted workflows at scale,” he says.
Vibe coding has a place, but it’s not the lead
Vibe coding, where natural language prompts generate code, has a role in this landscape. It’s useful for sketching ideas, exploring options, or building small apps or isolated prototypes. Vogels is clear that for simple, low impact applications, the risk is limited.
The problem begins when generated code becomes part of an existing system, particularly in regulated environments or infrastructure that affects security. Without insight into the broader context and a thorough review, no one truly knows what’s happening.
“Vibe coding without human oversight is gambling because you hope the generated code is correct, secure, and compliant, without truly verifying it.”
Governance protects long term maintainability
This is where governance becomes essential. AI-generated code can be clean and modern, yet still clash with existing architectures, naming conventions, or design principles. Boncode already sees the consequences in real projects that lack strong oversight.
AI-assisted coding works well for small isolated systems, but in business critical environments it requires much stronger governance. This is an investment in maintainability, ensuring that AI-assisted contributions align with architectural intent and organizational standards.
Code quality versus structural quality
One of the biggest impacts of AI-assisted development is the growing gap between local code quality and overall structural quality. Boncode assessments show that individual modules often improve when AI is used, while at the same time, the structure of the wider system can deteriorate
AI tools optimize for the immediate prompt, not the broader system context. Elegant components are injected into environments they were never designed to fit, increasing complexity and long term maintenance effort.
Human oversight remains essential
As AI takes on more coding tasks, human expertise shifts toward architectural thinking and system level decision-making. Like coaching a talented sports team, individual performance matters, but coordination and direction remain essential. AI can assist, but human oversight ensures readability, coherence, and long term value. In a world where developers spend most of their time reading code, that balance defines the future of software quality.
Boncode’s specialist tooling supports this approach by providing objective insight into both code quality and structural quality across the entire codebase. Jan Willem sums up the balance succinctly: “AI tools can be used to assist, but human oversight ensures long-term maintainability.” For organizations embracing AI-assisted coding, that balance is now one of the defining challenges of modern software development.
Get valuable insights into your software portfolio, with Boncode’s solutions for assessing and monitoring software quality.
