Baldur's given his thoughts on Bluesky - he suspects Zitron's downplayed some of AI's risks, chiefly in coding:

There’s even reason to believe that Ed’s downplaying some of the risks because they’re hard to quantify:

• The only plausible growth story today for the stock market as a whole is magical “AI” productivity growth. What happens to the market when that story fails?
• Coding isn’t the biggest “win” for LLMs but its biggest risk

Software dev has a bad habit of skipping research and design and just shipping poorly thought-out prototypes as products. These systems get increasingly harder to update over time and bugs proliferate. LLMs for coding magnify that risk.

We’re seeing companies ship software nobody in the company understands, with edge cases nobody is aware of, and a host of bugs. LLMs lead to code bases that are harder to understand, buggier, and much less secure.

LLMs for coding isn’t a productivity boon but the birth of a major Y2K-style crisis. Fixing Y2K cost the world’s economy over $500 billion USD (corrected for inflation), most of it borne by US institutions and companies.

And Y2K wasn’t promising magical growth on the order of trillions so the perceived loss of a failed AI Bubble in the eyes of the stock market would be much higher

On a related note, I suspect programming/software engineering's public image is going to spectacularly tank in the coming years - between the impending Y2K-style crisis Baldur points out, Silicon Valley going all-in on sucking up to Trump, and the myriad ways the slop-nami has hurt artists and non-artists alike, the pieces are in place to paint an image of programmers as incompetent fools at best and unrepentant fascists at worst.