The issue with AI is: it is only doing 80% of the work, and the easy 80%. You might know the pareto principle stating that 80% of an outcome comes from 20% of the work. Well, AI is doing the remaining 80%.

When using Claude, the AI writes boilerplate code and some tests, but the API design required careful thinking beforehand. The additional load the infrastructure will need to handle was evaluated for each possible solution. A prompt to steer Mr. AI in the correct direction and you are done. Yet, give the same tools to a junior without the experience needed and the results will be an unmaintainable spaghetti code that will crash your production infrastructure. 1000 LOC were written, 800 by the AI, 200 by the programmer, but 80% of the result is coming from these 200 lines. Once you know how to solve a problem, creating the solution is not hard.

Once a metric becomes a target, it stops being a good metric, don’t consider the 800 LOC as meaningful.