Glossary
The book’s terms, one line each. The development of each term lives in the chapter indicated. The terms the trilogy has already defined, such as the Nadella Line or Trust Infrastructure, keep their definition in the book that coined them; here the definition of those this book uses most is repeated, in one line.
software as we know it — The business frozen in a piece of software: described once, and changed only when someone opens a project. It makes no difference how it is held: rented as a service, licensed and operated on one’s own infrastructure or in one’s own cloud, or custom-built. It is what dies; not all software dies. (Chapter 1.)
second phase — The stretch of the crossing in which, with the business logic already in the agent layer, organizations start replacing their own underlying systems. The first phase collapses the interface; the second, the way of holding the software: the business described once and changed by project. (Chapter 1; the first phase, in AgencyDomains.)
agentic reading — The way of understanding “an agent builds and changes the system” in which the agent is a faster programmer: what comes out of it is an application that persists and that someone deploys and maintains. It takes its name from the agentic world of AgencyDomains. (Chapters 3 and 7.)
agentive reading — The way of understanding “an agent builds and changes the system” in which nobody programs it in advance: the agent attends to the work and crystallizes what repeats into pieces of its own, its muscle memory. It is this book’s. It takes its name from the agentive world of AgencyDomains. (Chapters 3 and 7.)
Botlet — A piece the agent executes from memory: code that resolves what repeats the same way every time, without consulting any model. It is the agent’s muscle memory and it is the agent’s own: it shapes it, maintains it and replaces it, although today the body of the starting piece is written by whoever maintains the catalog. The term is from AgencyDomains. (Chapter 3.)
Agentlet — The Botlet’s sibling piece for what repeats in its form but asks for judgment in every case: it carries a declared charter and a model sized to the task, not the largest one, with a budget per execution, and it escalates when it is not sure. It is also the agent’s. The term is from AgencyDomains. (Chapter 3.)
three-rung ladder — The way of ordering the agent’s work by what it costs to attend to it: the Botlet (code, no model), the Agentlet (a model sized to the task, with a budget) and full reasoning (the frontier model, reserved for what is new). It orders three things: cost, response time and dependency on outside vendors. It says where cost should come down, not by how much. (Chapter 3.)
Wingworking — The practice of work between a person and an artificial-intelligence agent with which this book was written: the person sets the problem, decides and judges; the agent proposes, drafts and verifies, like a wingman flying alongside the leader. AURA describes it as the step from using artificial intelligence as an assistant to delegating work to it. (Preface and colophon.)
wingcoding — Wingworking applied to building software: the person specifies and decides as an architect, and the agent programs. It is where the practice was born. (Chapter 1.)
harness — The environment an agent works in: the tools, the permissions, the memory and the rules that surround it while it attends to the work. (Chapter 1.)
template that comes to life — The starting point of a system of one’s own: it is installed whole and, from the first day, its owner evolves it with AI until it stops being the template. (Chapter 3.)
work that was never specified — The work that happens around the systems and that people today resolve by hand, because software only automates what someone specified in advance. (Chapters 1 and 3.)
bought peace of mind — What came included when buying software and nobody wrote into the budget: that many others used it, and that someone answered for it. It is noticed only when it is missing. (Chapter 4.)
system of record — The system that keeps the official version of a piece of data. It is what survives the crossing. (Chapters 4 and 5.)
system of truth — A system of record that also guarantees that what is recorded is true, through a contract with those who use it and with two guarantors. It is always a piece of technology, and it is operated by whoever uses it. (Chapter 5.)
two guarantors — Those who answer for the contract of a system of truth: the technology vendor, who answers for it meeting the six properties, and the one who attests, independent of whoever records, who verifies the records. With only the first, one is back to trusting the vendor; with only the second, what may have been overwritten is audited. (Chapter 5.)
truth contract — The written, verifiable and backed promise of a system of truth, in the technical sense of an agreement between parties. It promises six properties: nothing is deleted, it is corrected; the origin is traceable; it is verifiable by a third party; it is reconcilable; it speaks a common language; and it declares its degree of truth. (Chapter 5.)
degree of truth — The status of each entry in a system of truth: declared by someone, verified against its source or inferred. It is not a quality level: the required quality is a single one. (Chapter 5.)
the one who attests — The contract’s second guarantor: verifies the records independently of whoever records. It can be a human auditor or an automatic process; never the same organization that records. (Chapters 4 and 5.)
control layers — The five layers that make what is built with agents trustworthy, each with the question it answers: truth, identity and authority, movement, measurement and judgment. They are not the four layers of the AgencyDomains architecture. (Chapter 6.)
truth (layer) — The first control layer: the system of truth, with its two guarantors, for the records the organization is accountable to others for. It answers whether the numbers and the records are correct. (Chapter 6.)
movement (layer) — The third control layer: the agents and the platform they run on, with human supervision and an autonomy granted by degrees, process by process. (Chapter 6.)
measurement (layer) — The fourth control layer: every agent is tested against known cases before operating and keeps being measured while it operates. Trust in an agent is not declared: it is measured. (Chapter 6.)
judgment (layer) — The fifth control layer: industry standards as the starting point and an architecture team that decides what must exist and can judge what the agent shapes, even without reading the code. It is the defense against the dead end. (Chapter 6.)
people (dimension) — The condition that cuts across the five control layers: who operates each layer and governs its evolution, and with what training. (Chapter 6.)
reversal (dimension) — The condition that cuts across the five control layers: what happens when something fails and how to go back without stopping the business. It is not the personal reversibility of the agentive mode. (Chapter 6.)