There is no shortage of writing about the era of AI writing code. Much of it, though, is about how much productivity has gone up, and it sidesteps the question of what is actually left for developers. This series is a set of essays that takes that question head on.
Each part looks at one angle: whether it really doesn't matter that you can't read code as long as the spec is satisfied, what actually collapses and what survives when the boundaries between roles blur, and why the ability that becomes most expensive is the one that gets trained the least. To state the conclusion that runs through all three up front: production moves, but judgment and responsibility do not.
These are essays, not technical deep dives, so there are no measurements. Instead, each argument states the premises it relies on in the text, so if you disagree with something, it is worth starting with those premises.