Developers Shift 95% Test Coverage to LLMs as AI Takes Over Coding and Architecture
Updated
Updated · InfoWorld · Jul 22
Developers Shift 95% Test Coverage to LLMs as AI Takes Over Coding and Architecture
3 articles · Updated · InfoWorld · Jul 22
Summary
LLMs are increasingly handling core software work—coding, debugging, test generation and architecture—while developers move from manual implementation toward prompt-driven orchestration.
95% test coverage can now be produced with little effort, but the same model often writes both the application logic and the tests, raising the risk of self-reinforcing errors and weak validation.
Documentation, stack traces and deployment details are also being offloaded to AI, with developers pasting 200-line errors into chat tools and relying on generated schemas, security rules and infrastructure scripts they may barely review.
That shift is extending to planning and design, where AI-generated architecture documents and diagrams can impress managers even when underlying assumptions or scope flaws remain poorly understood.
The report argues this amounts to a new black-box phase in software engineering: developers still direct the outcome, but increasingly may not fully understand how the machine produced it.