AI

Suppose AI really does reach the heights we imagine for it, and society no longer needs the labour of most people — what are you going to do tomorrow morning? This piece asks one question from six directions: history, economics, anthropology, sociology, psychology and political philosophy. Are some of the predicaments we are trying to engineer away also the source of our social relationships and our sense of meaning? The core claim is that technical progress opens a space of possibilities, not a script written in advance; the central problem after AI is not how to produce more wealth but who owns the systems that produce it.

If AI eventually does most human intellectual work, one answer is soaring productivity, faster science and better medicine. Another is mass unemployment, concentrated wealth and unbalanced power. The two futures look contradictory and may both hold, because technical progress and social progress have never been the same thing. The industrial revolution raised output without abolishing poverty; the internet cut the cost of information without producing more reasonable public debate.

Society has long assumed knowledge is scarce, and built education, specialisation, consulting and investing on that assumption. AI is overturning it — but a falling cost of producing knowledge does not automatically raise the quality of human judgment. There is a worrying possibility instead: ever more powerful tools for processing information, alongside a fading ability to judge whether any of it is right. What is becoming scarce is epistemic judgment, not knowledge.

Two companies build AI agents. One invests in safety testing, restricts permissions and adds human review; the other shortens the cycle, opens up automation and undercuts on price. In the short run the aggressive one wins, so the careful one either loses ground or lowers its standards. Both end up aggressive. Nobody wanted an incident, and together they produced a market with a thinner safety margin. This requires no malice from anyone — only a competitive environment that cannot coordinate.

An AI that writes an investment memo and an AI that manages real money are in different risk classes — the first fails at the level of information, the second changes the state of real assets. Moving from generating content to taking action crosses an important line, and a great many products still evaluate autonomous systems with the methods built for content generators. Capability, goal alignment and permission to act have to be separated in the architecture.

Anthropic engineering manager Felix Rieseberg dropped the traditional coding test from his hiring loop, on the grounds that it no longer separates who will create value in the new environment. But the deeper question his piece points at is this: when knowledge is easy to get and ideas are easy to build, where does the difference between people come from? My answer is creativity, independent judgment, self-direction and responsibility — all four mattered before, but once execution stops being scarce they get reordered to the front.

Using the S-1 financials, long-term compute obligations, and Hacker News discussion described in the supplied material, this report distinguishes accounting losses, operating cash flow, and equity dilution, and examines cloud channels, governance, and 2030 valuation sensitivity. Offering information and original financial details still require direct review.

Rising Treasuries, falling technology stocks, pressured gold, and higher oil reflect intertwined growth concerns, AI financing risks, and liquidity-driven unwinds. Distinguishing short-term covering, recession pricing, and a stagflation reversal requires watching yield curves, credit spreads, and long-end auctions rather than a single trading day.