The scenes of urban warfare on Paris streets that have been prominent on X do not reflect reality when walking around central Paris. Cafes remained bustling, the Luxembourg Gardens were filled with people relaxing and even reading books, and fancy shops were open for business. Yet every now and again, you come across Paris antifa graffiti or a store with boards on the window with a sign assuring, “Votre magasin reste ouvert.”

The stores with boards were taking the downside bet that France is not done with its restiveness, a jumble of legitimate grievance, typical French activism and opportunistic thieves and rioters. The bet these shops were making is that disturbances are likely to return, even if the protests this past weekend passed without the kinds of scenes of violence and mayhem that led the reactionist types on X to proclaim “Paris is burning.” The Marais is still standing, even if the Hoka store is behind plywood barriers.

A safe bet, as America heads to the polls to vote today, is instability is a feature that will remain in place and likely even accelerate. Most of normal life continues, with punctuations of abnormality.

AI is a strange technology. It’s direction is uncertain. Anthropic can be predicted to be the largest company ever in history by orders of magnitude or go bust. Both are in play. Google has gone from an early AI winner with its pioneering work to a loser with the release of ChatGPT back to a winner thanks to its deft transition of its massive search ad business and now back to a loser as AI cuts into its search ads business and ChatGPT gains share. Meanwhile, the entire search ad market could get compressed since it was never actually generating demand but more of a tollbooth you put up when you have a monopoly as the interface to the world’s information. The commanding heights are great while you control them, but the come down is a drag. Also: search ads themselves are frequently friction. Be careful with holy wars.

Similarly, Reddit went from a winner as it became the citation source of choice and got one of the few AI licensing deals Google struck to seeing a sharp decline in traffic that’s caused it to overhaul its entire strategy by moving to a logged-in experience. RIP browsing and surfing. The open web will turn the information superhighway into a bunch of cul-de-sacs like you find in Phoenix.

Workers should by now have accepted instability as a feature. Companies will continue to refactor themselves. The gap has persisted between the type of productivity gains seen in tech – Anthropic expanded its revenue per employee more than quadrupled over the past year – and the rest of the economy. My read is that bosses will conclude that this isn’t a tech problem but a people problem. They’ll look to downsize to have leaner teams and fewer layers while AI systems take over much of coordination work and swathes of typical knowledge work. This will empower some and displace many others.

I’ve been immersed in making AI the center of The Rebooting’s business. I saw the light with how it can keep the company’s context and power much of our project delivery and commercial activity. I estimate about 15% of the company’s spending can be shifted from AI. The goal: Use the savings in coordination work and energy to build better products and serve clients better. I expect to pay far less for people to move data from one place to another or to monitor processes or update tracking documents. Some of that money spent will go to Anthropic and other AI companies, but much of it will go into growing the business. The reality in many businesses is they have very high coordination costs.

But the process is messy. I now have humans pointing out where AI has falled short and AI pointing at the same in humans. I’ve noticed AI has no chill. Claude will shift on a dime and overturn processes it just put in place minutes before. It will update wildly and always use 50 words when five will do. It can create chaos because it doesn’t think like a human or recognize how destabilizing constant change is.

Reminder: Join TRB Pro for access to our What’s Working case studies from leading publishing executives. Join now.

Even Dario Amodei is upfront that the greatest risk to the entire AI project is getting the timing wrong. “If I’m just off by a year in that rate of growth, or if the growth rate is 5x a year instead of 10x a year, then you go bankrupt,” Amodei told podcaster Dwarkesh Patel.

That’s an underlying sentiment in Anthropic’s S-1 filing, which shows a business that’s grown remarkably fast (revenue up up 1,300% in the first half of 2026) and, crucially, is showing how productive an AI company at scale can be ($6.6 million of revenue per employee).

The bull case hinges on the kinds of massive productivity gains seen in coding will be replicated in just about every facet of knowledge work. Anthropic’s $40 trillion total addressable market, as ludicrous as it sounds, hinges on it siphoning off a big chunk of several markets. David Sacks claims Amodei has talked about being the final company left. (Whether he ever said that is unproven.)

The bear case: These gains don’t materialize and AI becomes a commodity with token prices dropping while Anthropic’s infrastructure commitments remain locked in place. Creditors are starting to worry that the AI buildout is unsustainable by jacking up the borrowing costs that are underlying the buildout. A hallmark of bubbles is unsustainable debt to fuel capital expenditures. The railroads boom in the 1860s and 1870s was a classic productive bubble that led to a remade economy yet saw a financial crash because those gains didn’t materialize soon enough.

That is the concern today. joined me and Troy to discuss the Anthropic S-1 and this crucial underlying question that could determine how quickly AI permeates the economy.

“The No. 1 question in my mind is if the use cases for AI can keep growing at the same pace that the cost of the tokens are falling,” said CJ Gustafson from Mostly Metrics, who joined me and Troy Young for a conversation. “You can totally mistime this. They can actually work themselves into a situation where it’s not as economically viable to keep growing with the models they have.”

Listen to our conversation with CJ
Watch on YouTube