Skip to main content

Glossary

This page is generated from curriculum-state/ledgers/concept-ledger.yaml — specifically, from every entry whose status is defined. A term still marked used-undefined or proposed in that ledger is a known debt, not a settled definition, and does not appear here yet. If a later chapter reuses one of the words below, it reuses this exact wording — that consistency is the entire point of tracking terms in one place.

Agent​

A program that can read your files, write new code, edit existing code, and run commands on a computer — on its own, in a sequence, without asking you to approve every single step.

First met on the Welcome page. Put another way: a very fast, very capable helper — not an architect, and not the owner of the result.

Agent Skill​

A set of instructions kept in one folder that an AI agent reads when a task calls for it, so it handles that job the same way every time instead of improvising a new approach each attempt.

First met on the Welcome page. A skill is written once and reused, which is what separates it from a prompt you type fresh each time.

Agentic AI​

AI that does not only answer, but plans its steps and uses tools, such as searching, running a program or sending a message, to finish a job.

First met on the Welcome page. Building agentic AI of your own is one of the three skills this book teaches.

AI-native​

In this book, working with AI in four ways: using AI, building AI, building for AI (such as MCP servers), and adding AI to software that is not AI itself.

First met in the Thesis. Elsewhere the word often means something narrower, such as a company built around AI from its first day. This book uses the broader meaning.

Blast radius​

How much can break, and how far the damage reaches, if a change goes wrong.

First met in the Thesis.

Code Literacy​

The ability to read and judge code, whoever or whatever wrote it: reading the decisions and the evidence behind it, and the actual lines when needed, well enough to verify it and own it.

First met in the Thesis. This is the book's own name for its core skill, not a standard industry term. It is introduced in full the first time you meet it, and links back here after that.

Database​

An organised store of information that a program can search, add to and change, such as a shop's list of customers and their orders.

First met in the Thesis.

Framework​

A ready-made structure of code for building one kind of software, such as a website or a server, so you don't start from nothing. React and Django are well-known examples.

First met in the Thesis.

JavaScript​

The programming language that runs inside every web browser. It makes web pages interactive, and it can also run on servers.

First met in the Thesis.

MCP server​

A small program that gives an AI agent an ability it does not have on its own — reading your files, searching a database, sending a message — through one agreed connection. The agent asks, and the server does the work and answers.

First met on the Welcome page. MCP stands for Model Context Protocol, the standard that connection follows, so a tool written for one agent works with another.

Python​

A popular programming language known for its clear, readable style. It is widely used for AI, data work, automation and the servers behind websites.

First met on the Welcome page.

Software architecture​

The big decisions about how a piece of software is structured: its main parts, how they connect, and where its data lives. They are the decisions that are expensive to change later.

First met in the Thesis. Think of it as the house plan, drawn before the first brick.

Spec-driven development (SDD)​

The industry's practice of writing a specification before an AI writes any code, so the spec becomes the source of truth for both the person and the AI.

First met in the Thesis. Spec-Driven Engineering, this book's method, is built on it: SDD is about doing development with AI, and Spec-Driven Engineering is about doing the whole engineering job with AI.

Specification (spec)​

A written statement of what you want built and what shouldn't change, agreed before you start.

First met in the Thesis.

Spec-Driven Engineering​

The discipline in which the engineer specifies intent and boundaries, the AI implements within them, and a human verifies before taking ownership.

First met on the Welcome page. This is the book's own name for the idea at its centre, not a standard industry term. It is explained the first few times you meet it. After that, it links back to this page.

Technical debt​

The future cost of a shortcut taken today: work that has to be redone later, usually with extra effort, because something was built quickly or with only a partial understanding.

First met in the Thesis. The engineer Ward Cunningham coined the term, comparing first-draft code to borrowing money: a little debt speeds you up, as long as you pay it back soon.

Vibe coding​

Building software entirely through iterative AI prompts without understanding the underlying code. Coined by Andrej Karpathy, February 2025; named Collins Dictionary's Word of the Year for 2025.

First met in the Thesis. The term is not ours — it is credited to whoever coined it wherever it appears.

Vibe decisioning​

Making the decisions around software blindly, by feel or by accepting whatever the AI chose, without being able to make better ones than the AI. It matters most at the higher levels of engineering: AI agents, the cloud, and the structure of whole systems.

First met in the Thesis. I arrived at this term independently, and later learned that the analyst David Pidsley had used it first, in 2025, for building decision systems by prompting AI. This book uses it in the wider sense above.

On this page