A METR study has revealed a paradox in the work of modern developers: when using AI assistants, specialists subjectively feel a 20% acceleration in the process, yet actual productivity drops by 19%.
What Happened
According to data from METR, GitClear, and Faros AI, the use of generative AI leads to a shift in the workflow bottleneck. Instead of writing code, the primary burden becomes its verification. While the volume of generated code and the number of Pull Requests (PRs) are increasing, this is accompanied by an increase in Code Review time and a rise in the code churn rate (the need to rewrite code), which negates the potential benefits.
Context
The problem lies in a cognitive bias: AI significantly eases routine code generation tasks, creating a sense of "magic" and rapid progress. However, this masks the increasing complexity of debugging and verifying generated content, which requires deeper analysis and more time than traditional coding.
Why It Matters for the Industry
For the industry, this means the risk of making erroneous management decisions based on inadequate metrics, such as the number of commits or lines of code. An increase in code volume via AI is not equivalent to an increase in efficiency. Companies may need to revise their CI/CD processes and invest not in generation tools, but in automated verification tools and deep code quality auditing.
Why It Matters for Users
It is important for developers to realize that the feeling of speed from AI assistants can be deceptive. Instead of concentrating on the speed of writing code, professionals need to focus on the depth of verification and the ability to act as system architects and verifiers to avoid falling into the trap of fast but low-quality production.
Sources
Author
Look at AI, Editorial Team