Once AI is writing 30% of your code, your productivity *only* rises 2.4%!
That's WAY below what the optimistic RCTs would have suggested!
Quote
Johannes Wachs
@johannes_wachs
·
How much code now comes from AI? In new work with @simone_daniotti, @xiangnan_feng & @FrankNeffke we estimate that by end-2024 about 30% of Python functions pushed by US devs on GitHub are AI- generated. Adoption is rapid but diffusion lags globally. How did we do it?
Tbh writing code is only a small % of my time as a software engineer. 90% of my time is spent either debugging arcane company specific issues, integrating systems, or figuring out *what* code to write.
Writing code is just no longer the bottleneck - now I can handle 2-3x the number of projects and most time is spent orchestrating systems. It’s entirely possible this increased productivity is a net negative to my company by increasing organizational complexity
2.4% increase in the number of commits. I don't know one way or the other, but I wouldn't be surprised if a typical AI-generated commit does more than a human-generated one.
I wonder if there’s something related to adoption rates declining with experience that partially explains it, or perhaps those less likely to use AI might actually be better at goading the AI to do what they want. Or perhaps AI is changing task composition within individual jobs.
This reminds me of one of Isaac Asimov's short stories, The Last Question. Someone asks the computer about entropy. It can't come up with an answer. Eons later, when mankind is gone, the progeny of that computer learns how and implements it, "Let their be light!"
Press space bar to start a drag.
When dragging you can use the arrow keys to move the item around and escape to cancel.
Some screen readers may require you to be in focus mode or to use your pass through key