All of my X timeline is filled with posts like “wow, can’t believe how productive I am,” “haven’t shipped this much code since forever,” “I have 109 parallel agents working for me atm on different apps,” and “my productivity is now an order of magnitude higher”

Which naturally begs the question: if everyone is shipping so much more at an unprecedented rate, why aren’t we seeing more software out in the world?

If you look at the apps and services we use, there are no signs of faster iteration or fewer bugs since AI started taking our jobs. This lack of impact from AI coding tools on actual software quality and delivery speed has been very amusing to watch unfold in real time.

First, there’s a Cantillon effect at play here. Developers are closest to the spigot, so most software being built right now is meta software, tools for devs by devs. They are building for themselves, obviously, that’s their user problem, the world they know. They are not aware (nor do they care) of problems outside their bubble, so they’re not building to solve those. The productivity gains haven’t reached other domains yet because they’re stuck in this loop.

Second, most developers are sudoku solvers, they really enjoy the logic puzzles and navigating the complexity of their own custom built mazes, and hoarding the context and intricacies of these systems to themselves, while only passing it down to those they mentor and deem as worthy. But without a clear product direction, most of the time they’re just shuffling code around from one place to another, creating technical todo lists to check off without ever reaching an end product. In times of old, software companies would have tolerated and even bowed down to this kind of attitude, but now that moat is drying up, they have no leverage.

It’s the George Costanza problem, a fundamental lack of empathy. If you can’t think in the user’s terms or figure out their problems in the first place, AI just helps you build without purpose. Previously, this was seen as having some value because look at all the effort and time you put into this, this is surely worth something, we must use this somewhere now is just a couple prompts away. We’ve reached the ‘tree falling in the forest’ stage of development, if the agents generate code that solves zero user problems, has anything actually been built? You won’t ever fix the world, even with the H2G2-supercomputer at your disposal, you will be confined to (and content with) playing in the sandbox.