The promises have been bold: AI coding tools will revolutionise software development, slash delivery times, and turn every developer into a 10x engineer. But what actually happens when software developers use these tools in practice?
At Codurance, we decided to move beyond the marketing claims and find out. We brought together senior software engineers for hands-on coding challenges. Half of them using AI assistance tools like GitHub Copilot, Cursor, and Amazon Q, and half coding without any AI support. Then we analysed the code they produced. The results? More complex than the hype suggests.
Our study revealed some eye-opening findings that every technical leader should understand:
AI generated code tended to be larger, much larger. AI-assisted solutions contained an average of 4x more lines of code than their non-AI counterparts, despite having the same time constraints. While teams accomplished more features, the codebase bloat was significant.
Complexity Increases. Cyclomatic complexity, a key measure of code maintainability, was almost 4x higher in AI-generated code. This means more branching logic, more conditional paths, and ultimately, automated tests needed to safeguard against the increased opportunities for defects.
The productivity paradox. Every developer reported feeling incredibly productive at first. Initial setup was lightning-fast, boilerplate vanished, and progress felt effortless. But this productivity boost plateaued quickly. As one group put it, despite their codebase being only hours old, "it already felt like working with legacy code."
Perhaps most concerning was how developers described losing control of their own code. The AI tools were "running away with them," generating code faster than they could meaningfully review. Without careful discipline, the AI started driving architectural decisions rather than simply assisting with implementation.
This isn't just a productivity issue, it's a strategic risk. When AI tools make design decisions optimised for speed rather than maintainability, you're trading short-term velocity for long-term technical debt.
AI coding tools aren't the silver bullet they are often marketed as, but at the same time they are not without value. They excel at prototyping, scaffolding, and rapid exploration, all scenarios where speed matters more than long-term maintainability.
For enterprise development, success requires:
The 2024 State of DevOps report already warned that large batches of AI-generated code can harm overall team productivity. Our findings support this: without proper governance, AI tools can create more problems than they solve.
AI tools offer genuine opportunities across multiple areas of software development, from accelerating prototyping and eliminating the overhead of boilerplate code to speeding up initial development phases. But strategic adoption is equally critical. The teams that will thrive are those who treat AI not as a shortcut, but as a powerful collaborator that requires skill, discipline, and strategic thinking to use effectively.
Ready to ensure your team's AI adoption drives real value without compromising code quality? Download the full report for detailed recommendations and methodology, or reach out to discuss how we can help assess your team's readiness for AI coding tools.
The future of software development isn't about replacing developers with AI, it's about empowering developers to stay firmly in control while leveraging AI's strengths.
If you are looking to get started on your AI journey or would like some support to power your AI adoption strategy then get in touch, Codurance would love to help you realise the full potential of AI.