AI coding got faster. Why didn’t engineering?

AI coding tools have become faster, but engineering organizations are not seeing increased productivity or innovation as a result. This is largely due to the cost of AI investment not being offset by the velocity of engineering teams. Despite a 28-fold increase in AI investment, particularly for larger engineering organizations with over 99 engineers, measures of velocity remain stagnant or even declining.

The State of AI Impact in Engineering report from DX found that while AI spends continue to rise, the allocation of engineering effort towards new feature work, as opposed to maintenance and operational overhead, remains flat. This means that AI is not effectively freeing up engineers to focus on business solutions.

One possible explanation for this disconnect is that the time saved by AI tools is being spent on addressing existing technical debt and backlog items, rather than delivering new features. Additionally, there is growing concern about developer experience, particularly around code maintainability and change confidence. AI makes it easier to understand and modify code, but engineers are now more hesitant to release changes, leading to a decrease in change confidence.

Smaller organizations seem to benefit more from AI investment, likely due to having less overhead in terms of communication and coordination. As organizations grow larger, the coordination costs associated with AI implementation increase, making it harder to see returns on investment. This is especially true for legacy software organizations, which struggle to adapt to the fundamental shift that AI represents.

Some experts argue that companies stuck with outdated structures and processes are unable to fully benefit from AI, likening their situation to retrofitting a building that was built when code was expensive. Instead, they propose that companies start fresh with new, AI-informed structures.