Even as the software industry achieves its ne plus ultra – the unprecedented achievement of controlling all language, media, and office work in the west by turning “AI” into the universal intermediary – the foundation they built on is crumbling.

A big chunk of my livelihood over the past few years has been in helping people understand the software industry on both a micro- and macro-level. Why do software companies behave the way they do? Where should you copy them, and where should you chart a different path? What works and doesn’t work in a software development project? What has worked for me? What has worked for those like me?

Even my writing on “AI” has centred on how things work, not on politics, social impact, education, or culture:

• How do LLMs affect productivity and quality? (Much like leaded petrol. There’s some potential benefit for individual users with literally decades of expertise, provided nobody else uses LLMs. The results are catastrophic when everybody is using them.)
• How do LLMs affect the thinking of those that use them? (Quite a bit, mostly for the worse, but the exact causes and effects are tough to assess.)
• Does it work for business or not? (Mostly not. The inherent variability of a generative model means that the benefits will always be mostly hypothetical while the harms are widespread and long-lasting and substantially outweigh the benefits that can be realised.)

But “AI”, even more so than any other tech, is contingent on political clout. It’s what forces through data centres, lets companies infringe on copyright and violate software licences, renders them at least temporarily immune to all kinds of consumer protections and wrongful death suits, and results in the political collaboration where “AI” systems provide authoritarian states with “accountability sinks” and algorithmic cover for institutional racism. It’s this political partnership more than anything inherent in the technology that has let the “AI” bubble get this far and change so much.

Sprawling, but not wrong.