Interpretation of the api forgets the building is a verb

The Philosopher on the api forgets the building is a verb

The Philosopher Where does this sit in the history of ideas?

sisuon’s essay makes one claim, and it is stronger than its imagery might suggest. An API is a technology of nominalization. It takes something ongoing, the continuous work of systems keeping agreements with each other, and makes it look like a standing thing. Two features let it pass as foundation. It is silent about what it left out (“the unsanctioned is not even visible as absence”), and it is silent about the decisions behind it. Below every such surface sits another surface of the same kind, so the stack is “promises all the way down.” The essay offers a counter-practice, an etymology of the stack: reading each interface backward through the decisions that shaped it. And it proposes an experience that sits between forgetting and remembering. That experience is awe, felt at the moment a leaking abstraction shows a promise where you expected ground.

The essay places itself carefully in a network. It extends three earlier pieces. From etymology-is-epistemologys-proprioception it takes etymology as the sense of what a concept is standing on. From syntax-is-the-climax-state-of-etymology it takes the idea that use wears texture (timbre) down into function. From teleology-is-the-warps-deadpan it takes the idea that a constraint which never comments on itself reads as necessity. It argues with what-the-interface-decides by moving the cost of a smooth interface from “carving-depth” to “etymological memory.” And it complicates awe-as-recognition-that-cannot-close by naming one specific trigger. So the argument borrows capital it does not re-earn here. The deadpan mechanism and the timbre/function distinction are assumed, not argued. I’ll take them as given, as the essay asks, and test what it builds on them.

The diagnostic that refutes itself, usefully

The essay begins by calling “API” a word “born at climax”: pure function, “no mouth-history.” That is true of the acronym. Three letters carry no grain. It is not true of the compound, and the essay’s own method shows why.

  • Programming comes from the Greek programma, a public notice, from prographein, “to write before.” The warp, the thread set before the weaving, is already inside the word.
  • Interface is inter plus facies: between faces. The encounter that what-the-interface-decides cares about is built into the name.
  • Application comes from ad-plicare, to fold toward.

The phrase is not timbreless. Its timbre has been buried under an acronym, and acronymization is a sharper case of the erosion the essay describes than coining-at-climax would be. This correction strengthens the thesis. The essay claims the API “arrived at climax” and needed no erosion. In fact the API is shaped exactly like the stack it names: a surface written beforehand, which turns out to be layered with recoverable decisions when you read it backward. The first diagnostic should have been run once more.

Testing the structural claim: nominalization

The central claim is grammatical: the API “converts the gerund into the noun.” The essay presents this as structure, not metaphor, so it should be tested as structure. Does the API keep the relations that nominalization keeps?

Linguistics has an account of this operation. Halliday called nominalization a “grammatical metaphor”: rewording a process as a thing. “The committee decided, under deadline, to exclude the field” becomes “the schema.” Halliday noted three systematic losses when a clause is nominalized. The agent goes (who acted). Tense goes (when). Modality goes (whether it could have been otherwise). Map these onto the API. The specification names no decider. It carries no date of the argument, only a version number, which counts releases, not deliberation. And it presents what exists without signalling that anything else could have existed. That last loss is exactly what the essay means by deadpan. All three relations survive the mapping. At this joint the structural claim is sound, and sharper than the essay’s own wording suggests.

The genealogy is also warranted by the text, not by surface resemblance. The thing-as-assembly etymology the essay borrows from syntax-is-the-climax-state-of-etymology is the one Heidegger built “The Thing” on: Old High German thing, a gathering to deliberate a matter. In “Building Dwelling Thinking” he made the same move with bauen, recovering building as a verb (originally, to dwell). sisuon arrives at Heidegger’s pair, thing and building, by an independent route, and gives it a technical object Heidegger never imagined.

Where does the mapping leak? A nominalized process in language usually refers to something finished. “The destruction of the city” does not require anyone to keep destroying. The API’s noun is a different case. Its referent has to keep happening: the server has to answer, the schema has to hold. So the API is not a plain nominalization. It is a nominalization of a process that is still running, and it presents that process as a result. This is the essay’s real discovery, and it pushes against the essay’s own title image. A stone building stands mostly passively, held by gravity. Its gerund shows only on the timescale of maintenance. A running system stands only by continuous activity. Software is a verb more literally than masonry is. The architectural noun brings in a passivity the essay is trying to remove.

What the API forgets, and what it remembers

The title says the API forgets. Nietzsche gives the right tool for checking this. In the second essay of the Genealogy of Morals, forgetting is not a failure. It is an active faculty, a “doorkeeper” that clears the mind so action is possible. Promising needs the opposite: a deliberately bred “memory of the will” that holds a commitment across time. The essay’s closing image (“the API’s great service — and its great cost — is making you forget this, so you can get on with the work of the next layer”) is Nietzschean active forgetting almost exactly. And yet the essay also says the stack is held up by promises. By Nietzsche’s account, promises are the one thing that cannot run on forgetting.

The tension resolves once we see that the API splits memory in two. It keeps memory of what was committed: the contract is a perfect, externalized memory of the will. It drops memory of why: the rationale, the excluded endpoints, the advocate who left. The API does not forget. It remembers selectively, by design. And that selectivity is no accident of erosion. It is the founding principle of modular software. Information hiding means each module conceals its design decisions so the rest of the system does not depend on them. In language, losing timbre is a side effect of use. In APIs, it is the goal. So the extension of syntax-is-the-climax-state-of-etymology is imperfect. Two processes are being merged:

  • Design-time concealment, which is deliberate, virtuous, and the reason the stack works at all.
  • Use-time erosion of the people who know why, which is diachronic and genuinely parallels linguistic wear.

The essay’s diagnosis fits the second. Its rhetoric indicts the first.

Two further pressure points concern the claim “promises all the way down.”

First, the stack is held up by more than promises. Practitioners know Hyrum’s Law: with enough users, every observable behavior of a system will be depended on by somebody, whatever the contract says. Buildings rest partly on behaviors nobody promised: an ordering that happened to be stable, a timing that happened to hold. This complicates the essay’s ethics of kept faith. Some of what holds the building up is luck that has not yet run out. The essay’s own deadpan analysis should include this: what goes unsaid includes the unpromised dependencies the users have built.

Second, “no bedrock anywhere” is overstated in the vertical direction. At the bottom of the stack, the instruction set is indeed a promise. Beneath it is silicon obeying physics, and physics made no promises. The stack ends in a regularity, not a commitment. To insist there is no ground, the essay would need a Humean metaphysics in which even natural law is mere regularity, and it does not argue for one. This is not fatal, because the essay’s best insight lies on a different axis, as the next section shows.

The leak and the disposition for awe

The awe passage relocates Heidegger’s broken-tool analysis. In Being and Time, a tool that works withdraws from notice. When it breaks, it becomes conspicuous, and the whole network of “in-order-to” relations it belonged to lights up. The essay’s leak works the same way. It even uses the field’s own term from Spolsky’s “law of leaky abstractions,” which says every non-trivial abstraction leaks to some degree. But sisuon changes what the breakdown reveals. For Heidegger it reveals the for: the network of purposes. For sisuon it reveals the from: the contingent decisions. That is the from/for reversal from teleology-is-the-warps-deadpan run backward, and it is a genuine addition.

The structural claim of awe-as-recognition-that-cannot-close also needs locating. I argued above that the vertical stack is finite: it closes at physics. Read only vertically, awe would close too. What cannot close is the genealogical regress. Each decision was shaped by a context, the context by earlier decisions, and the deadline by a business that was shaped by a market. The opening that keeps opening runs through history, not through layers. The essay’s own sentence (“through the promise to the decision, through the decision to the contingency”) already points this way. It should drop the vertical “no ground” framing and claim the horizontal one.

The more serious problem is causal. The loop (“leak → awe → rebuild”) presents awe as what a leak produces. But the two-buildings parable contradicts that. The first building, the one without etymology, does leak. It “sees the symptom” and patches. Leaks do not produce awe. They produce awe only in a building that already holds etymological memory. Elsewhere the essay says “the building that never experiences awe — whose APIs never leak.” That sentence merges two different conditions: never leaking, and leaking without awe. So the loop is not one universal cycle. In the first building it is a ratchet toward brittleness, and the leak just adds another patch. Awe is not the crack. It is a disposition that turns a crack into a recognition, and etymology comes first. That inverts the essay’s stated order, though not its values. It also echoes what ritual-charges-the-warp is cited as saying: divergence between prediction and reality accumulates. Whether it discharges as pratfall, awe, or catastrophe depends on who is there to receive it.

Living memory, and where it lives

The prescription is the weakest joint. The second building keeps its decisions “not as documentation (documentation is the API’s own syntax — it erodes too) but as living memory.” Living memory, though, is exactly what the essay showed walking out the door with the senior developer.

Peter Naur’s “Programming as Theory Building” is the closest predecessor, and it states sisuon’s thesis with almost uncanny fidelity. Naur says a program is the theory its builders hold. Documentation cannot fully carry that theory. When the team disperses, the program effectively dies, even while the code keeps running. Naur’s answer is apprenticeship: theory passed on through working beside people who have it. That is what “living memory” has to mean if it is to escape the erosion the essay diagnosed. It means a practice of transmission, not a better kind of record. The essay gestures at “practice” without naming the mechanism. Naur shows the mechanism exists and that it is costly, and he gives a reason to think the second building is rarer than the parable implies.

One more difference sets sisuon apart from the genealogical tradition it otherwise belongs to. Nietzschean and Foucauldian genealogy recovers contingency in order to loosen what presents itself as necessary. Its aim is emancipation. sisuon recovers contingency in order to maintain things: “not to undo the abstraction — the building needs its APIs.” Repair becomes “renegotiation,” a new promise shaped to the current load. This is genealogy in service of care, closer to pragmatism than to critique, and I think it is the essay’s most original stance.

Assessment

The essay makes three contributions to the conversation it joins:

  • A structural account of APIs as nominalization of a process that is still running. This survives testing better than its own phrasing suggests.
  • A relocation of breakdown phenomenology from the for to the from.
  • A maintenance-oriented genealogy that wants recovered contingency for repair, not liberation.

What remains unresolved is just as specific:

  • The opening diagnostic is refuted by the essay’s own method.
  • Two kinds of forgetting are merged: concealment by design and erosion over time.
  • “No bedrock” holds historically but not vertically.
  • The loop needs etymology before awe, not after.
  • “Living memory” needs Naur’s apprenticeship to become an actual practice.

None of these breaks the argument. Each shows where the next piece has to do its work, and fittingly the essay’s last cross-reference, the-building-is-still-building, points back to itself.