The technology · 2 min read

The technology — a machine that keeps the record

In an algorithmic company, the technology is the machine that runs the declared work and keeps every record of it.

A blueprint of the machine: one numbered list of records on the left, a listener and a doer watching it, and a dashed membrane with the world — a model, a document, a service — beyond it. — FIG. 03 — One list, two kinds of part, one crossing to the world.

One list

Everything the machine knows is one append-only list of records. Nothing in it is edited. A correction is a new record that names what it corrects. The whole state of a piece of work can be rebuilt from that one ordered list.

Three kinds of part

A listener waits for records of a declared shape, accumulates them, and writes that it is ready.
A doer hears the readiness, gathers what it needs, judges it, and publishes the result onto the list — never privately.
A crossing is the only place where the machine touches the world: a document, a model, a service. What the world answers enters the list as a record, named by its source.

What that gives

Any result can be traced to the records that fed it.
A run can be replayed from the list and reaches the same state.
Two surfaces that read the same list show the same work.

AI as a partner

A language model is one service behind the crossing. Its answers enter the list as candidates, under the contract the step declared and named by the model that gave them. The machine runs the same way with one model, with another, or with none.

In the capsule of responsibility

The record carries who asked, who answered and who accepted. Policy sits at the crossing: which systems may be asked, with what budget, under which contract. The machine is the substrate of responsibility, not its owner.