An AI agent speeds work equally well on both healthy and poor architecture - the difference is what it speeds up. GitClear research (211 million lines of changes) recorded an eightfold rise in blocks with five or more duplicated lines in 2024; copy-paste rose from 8.3% to 12.3% of all changes, while refactoring fell from 25% to under 10%.
Practical takeaway: the agent tends to generate working snippets and copy them across modules without consolidating them. This creates a layer of technical debt that grows quietly and later shifts the team’s time from building new things to fixing old ones. That is why the AI-native approach works only on top of an engineering foundation, such as loose coupling, reusable components, and build checks, not instead of it.
