Chapter 22. The Agent Era

Improper architecture is locally cheap and globally expensive: fusing is always less work today, and the costs land on caches, crawlers, integrators, and the future. For thirty years the future could wait. Now the bill arrives: software agents are trying to read the web, and they find what Part IV measured — rendered pixels and private APIs. The response is a compensating industry assembling itself in real time — scraping harnesses, headless browsers, and a per-application protocol server bolted onto every system that wants to be machine-readable. Read that list against Chapter 11’s scores: it is the S4 tax, collected one adapter at a time, at industry scale, exactly as the model predicts. The machine-readable web is being retrofitted at the margin because it lapsed at the core — Chapter 9’s maintenance failure, its cost still compounding.

On the left, four agents each wired to three applications through a dense tangle of one-off adapters, labelled N times M; on the right, the same agents and applications each connecting once to a single shared substrate, labelled N plus M

The arithmetic of that compensating industry is the integration industry’s arithmetic (Chapter 17) at a new scale. N agents meeting M applications through bespoke adapters need on the order of N × M integrations. The moment state shares one model and one query semantics, the count collapses to N + M — each side implements the common substrate once. Every generation of middleware has re-learned this sum. The agent era re-learns it with N growing by the month. The per-application protocol server (MCP, the emerging convention as of this writing) is the N × M answer, shipped in real time. The protocol is shared, but the model and query semantics are not: each server exposes its own vocabulary, so every agent still learns every application one at a time. The derived stack is the N + M answer, shipped since 1999.

There is an exit built into this arithmetic. An adapter that translates one application into the common model is written once, for everyone — not once per consumer, forever. It is the rare piece of glue that retires on success: the day the application serves its own state, the adapter’s answers and the application’s coincide, and nothing downstream notices the handoff. Until then the adapter offers a choice the silo never did. Its answers can stay virtual — computed from the silo at the moment they are asked, held nowhere. Or they can be drawn through the same read into state under the querying party’s own origin, where they outlive the silo account they came from. The industry already builds compensating machinery around every silo. Point it at the common model, once per silo, and that machinery is the bridge.

The vision, dated 2001. The scenario this chapter derives was written as fiction twenty-five years ago. The May 2001 Scientific American article “The Semantic Web” is by Tim Berners-Lee, James Hendler, and Ora Lassila. It opens with Lucy’s agent negotiating a course of medical appointments over machine-readable data on her behalf — an agent reading the web, not scraping its pixels. It read as science fiction because the agents did not exist. They exist now. Chapter 8 said the substrate was built for machine consumption and the machines were twenty years out; this is the same clock, run down. What was missing was never the stack — it was the reader, and the reader has arrived.

Underneath runs the grounding problem. Statistical models interpolate, and interpolation hallucinates. What agents need beneath them is a substrate whose answers are computed rather than guessed. Fact-sets with a formal query semantics are that substrate, and Part III named the deployed one. The hybrid of the two — statistical model above, fact-set substrate below — has a specific shape. The model interprets the long tail — the rare, one-off cases the substrate does not yet cover. What proves valuable there is promoted into the substrate, and promotion is nothing exotic: it is a write, an assertion made under the origin of a party that vouches for it. The substrate is governed as any origin is — by whoever holds the right to write it. And integration accumulates instead of repeating: every promoted fact composes by union, and stays.

“Machine-consumable” unpacks to nothing new: state with a universal model (R1), a coordination-free merge (R2), global reference (R3), and addressable intermediates (S4). An agent is a user agent. The requirement was sitting in Definition 1.1’s type signature all along.

And reading is half of Definition 1.1; the write side serves agents twice over. An agent’s change arrives as Chapter 7’s delta: two fact-sets, (D⁻, D⁺). The delta is a reviewable object: a human can inspect it before it applies, an audit log can store it verbatim, an operator can invert it by swapping the sets. Compare the alternative on offer: an opaque API call whose effect is whatever the endpoint’s code decided, reversible by nothing. Agent autonomy is a governance problem exactly as far as agent actions are opaque, and the delta normal form makes the action a document. And what holds for one change holds for a whole plan of them. An agent’s entire intended course can be stated as one document and read before any of it runs — what it will read, what it will change, what it will do if the first answer disappoints. The agent is audited not by trusting its account of itself afterward, but by reading its plan before. Chapter 7 derived it for humans holding forms; it turns out to have been waiting for machines.

The chapter ends with a question, put to any agent directly: is it more efficient for you to write a custom system for every domain, or to reuse one generic system and define the domain as data? The answer is not in doubt. What stands between agents and the second option is the set of human mental models Part IV audited — pre-web paradigms, defended now by habit rather than argument. The web that agents need is the web this book derived — necessarily. Both derive from the same requirement: machine-consumable, mergeable, globally referenced state.