The whole shift is that it questions what work “can’t be automated or accelerated”. Not denying Amdahl’s Law, but I think the primary unknown is how it applies.
I suspect the issue is fractal—it’s not about what percentage of tasks can be automated, it’s about what fraction of meta- and meta-meta- (and so on) tasks can be automated. How much human input and supervision is needed to come up with goals, break down into projects, sub-projects, tasks, sub-tasks, etc. And to supervise and deal with discoveries that the initial desire wasn’t quite what they want and adjustments are needed.
I have seen LLMs and their tooling move up the tree over the last few years VERY rapidly. I don’t have much faith in any prediction of future path—maybe there’re parts a LOT harder than we realize. Maybe it’s just context and speed.
The whole shift is that it questions what work “can’t be automated or accelerated”. Not denying Amdahl’s Law, but I think the primary unknown is how it applies.
I suspect the issue is fractal—it’s not about what percentage of tasks can be automated, it’s about what fraction of meta- and meta-meta- (and so on) tasks can be automated. How much human input and supervision is needed to come up with goals, break down into projects, sub-projects, tasks, sub-tasks, etc. And to supervise and deal with discoveries that the initial desire wasn’t quite what they want and adjustments are needed.
I have seen LLMs and their tooling move up the tree over the last few years VERY rapidly. I don’t have much faith in any prediction of future path—maybe there’re parts a LOT harder than we realize. Maybe it’s just context and speed.