In earlier days, we didn't want to spend our attention on machine code or assembly, and later memory management an garbage collection. So we invented tools and abstractions so that we didn't have to think about these things any more. We had more valuable problems and since we could automate them away, they were no longer deemed worth our attention.

The same is now happening with high level code. We're spending less and less time with the code, both writing and understanding it because AI / LLMs / Agents can automate much of the management of code, and anyway, we have more valuable problems.

The difference I think is that while earlier solutions to more fundamental computing concerns were deterministic processes, AI is probabilistic. So while we can outsource our attention for a wider range of problems, we're going to be pulled back in for decision making, checking and 'taste' tests more regularly than we would have been previously, and this will happen alongside rapid proliferation of new code being written and deployed.

I'm not yet sure what the full consequence of this will be, but I feel it's what is at the core of widely reported developer fatigue - the demand for attention is not going away, and the range of problems that require that attention will also grow.

I think we can summarise that we are just doing more, faster. But not all resources (such as attention and the earth's physical resources are going to keep up).