If you have a tool which does a good job at automatically generating a certain type of code, you should probably generate more of that type than you used to. This could easily balloon to 90% of your new code, but only replace a small amount of human coding. You seem to allude to this scenario when you talk about one-off scripts. Another example is automated tests.
This. LLMs make it much easier to write one-off scripts and custom tools for a bunch of stuff like plotting and visualizations. Because the cost is lower you produce much more of this stuff. But this doesn’t mean that AI is contributing 90% of the value. The ideal metric would be something like “how much longer would it take to ship Claude n+1 if we didn’t use Claude n for coding”?
The denominator isn’t fixed.
If you have a tool which does a good job at automatically generating a certain type of code, you should probably generate more of that type than you used to. This could easily balloon to 90% of your new code, but only replace a small amount of human coding. You seem to allude to this scenario when you talk about one-off scripts. Another example is automated tests.
This. LLMs make it much easier to write one-off scripts and custom tools for a bunch of stuff like plotting and visualizations. Because the cost is lower you produce much more of this stuff. But this doesn’t mean that AI is contributing 90% of the value. The ideal metric would be something like “how much longer would it take to ship Claude n+1 if we didn’t use Claude n for coding”?