Part IV

Capital has long embodied previous work. Computational capital adds a disturbing possibility: parts of the productive apparatus can now search, recommend, initiate bounded actions, and assist in the design of successor systems. A machine need not own itself or possess a legal personality for allocation to begin passing through it.

That passage has to be described carefully. Selection is not ownership. Prediction is not authority. A runtime may receive discretion over a case without becoming the institution that authorized its action or the principal that survives to answer for it. Recursive production likewise admits degrees: systems already assist coding, evaluation, experimentation, and chip design, while objectives, fabs, power, capital, and institutional permission remain external constraints.

Deployment therefore depends on more than capability. Verification cost is one candidate ordering variable, but integration, expected loss, liability, complementary labor, regulation, demand, and organizational capacity can dominate it. High-value work that is cheap to check may automate early. So may low-verifiability work when institutions tolerate the risk. The ratio developed in Chapter 16 is a heuristic for finding that pressure, not a law of succession.

Part IV follows capital into that incomplete loop. It asks what can be delegated, which decisions remain with a principal, how collateral can enforce bounded promises, where actuation interrupts software's apparent autonomy, and what survives the counter-thesis. The question is not whether machines have become economic persons. It is how much economic selection can occur before any person is summoned to answer.