I do a lot of AI-assisted coding now and I think this is right, but the timelines are overstated. There are still problems that tilt the risk/reward to the John Henry side.
I do a lot of AI-assisted coding now and I think this is right, but the timelines are overstated. There are still problems that tilt the risk/reward to the John Henry side.
Any useful codebase is going to be of a certain size, have defined coding/testing/org standards, have business requirement contradictions, etc. and those things produce gigantic context windows and can lead to generative messes that are expensive to clean up.