Glossary
Agency
The capacity to choose a direction and act on it. In this book, agency means deciding what you want to make possible, directing the work, and remaining responsible for your choices.
AI agent
A system that uses an AI model and tools to pursue a goal across multiple steps. Its ability to act depends on the tools, permissions, context, and oversight it has been given.
AI fluency
The practical ability to choose an approach, give AI useful direction, evaluate the result, and adapt what you learn to another task. It develops through practice and judgment, beyond familiarity with a particular tool.
AI Fluency Loop
This book’s experimental framework: define your purpose, choose your approach, experiment with AI, check and improve, and apply what you learned. Move between the steps as the work requires.
AI literacy
The knowledge and critical understanding needed to make sense of AI: what it can do, how it can fail, how it affects people, and when its use is appropriate.
AI-native technology stack
A connected set of capabilities used to investigate, plan, make, and revise with AI throughout a project. Shared briefs, evidence, files, and decisions help the work travel between tools.
Artificial general imagination
A phrase used in this book to describe the creative possibilities that emerge when AI capabilities across media can be combined. It is an interpretive idea, not an established technical category or a claim that every imagined work is achievable.
Artificial general intelligence (AGI)
A contested term for AI with broad capabilities across many kinds of intellectual work. Definitions differ about the required breadth, performance, autonomy, and ability to learn; a declaration of AGI does not establish consensus.
Artificial intelligence (AI)
A broad field concerned with computational systems that perform tasks associated with intelligence, such as recognizing patterns, generating language, making predictions, or planning actions. Different AI systems have different capabilities.
Artificial superintelligence (ASI)
AI that would substantially exceed human capabilities across a broad range of important intellectual tasks. Discussions of ASI concern a possible level of capability, not a settled milestone or a single agreed test.
Automation
Having a system carry out a task or process with reduced human intervention. Automating production does not remove the need to decide whether the task, result, or consequences are acceptable.
Bias
Systematic tendencies that can skew a system’s outputs or affect groups unevenly. Bias can arise from data, design choices, evaluation, or the circumstances in which a system is used.
Brief
A useful account of what the work should accomplish, whom it serves, what material and constraints matter, and how the result will be judged. A brief can guide several prompts and several tools.
Chatbot
A system people interact with through conversation. A chatbot may generate responses alone or connect to tools for searching, creating files, or taking actions; the conversational interface does not tell you all of its capabilities.
Coding agent
An AI agent equipped to work on software tasks, such as editing files, running code, checking results, and revising an implementation. A working preview still needs evaluation against the intended purpose.
Cognitive offloading
Using an external aid to reduce the mental work a task requires. In learning, the important question is which effort is being delegated and whether the learner still develops the intended understanding and skill.
Connector
An integration that lets an AI system access information or tools in another service. What it can read or change depends on the integration and the permissions granted.
Context window
The amount of information a model can handle in a single interaction, usually measured in tokens. A large window does not guarantee that every detail will be used correctly or retained in later work.
Design literacy
The ability to understand people’s needs, shape a purposeful response, and judge whether it works. In this book, it connects research, clear direction, making, and feedback from the people a creation is meant to serve.
Digital divide
Unequal access to digital technologies and the resources and skills needed to use them. The book’s intelligence gap extends that concern to access to capable AI and opportunities to learn, imagine, and act with it.
Frontier model
A model near the leading edge of capability at a particular time. The term is relative: the frontier changes, and strength on one task does not establish equal strength on every task.
Generative AI
AI that produces new outputs, such as text, images, audio, video, or code, from patterns learned during training and the input it receives. A generated result may still be inaccurate, derivative, or unsuitable for its purpose.
Grounding
Connecting an AI response to relevant material, such as supplied documents, retrieved sources, or observed data. Grounding can support a more informed response, but readers still need to check whether the sources support the claims.
Hallucination
A plausible-sounding but incorrect or unsupported AI output. Examples include invented sources, quotations, or details. Confidence and fluency in a response are not evidence that it is true.
Imagination
The capacity to envision what could exist and explore alternatives. Here it means seeing possibilities for a story, lesson, tool, or experience and considering whom those possibilities might help.
Imagination Age
An idea with a history before generative AI that emphasizes imagination and creative expression as defining capacities. This book uses it to explore how expanding creative tools change who can make things and what they can attempt.
Intelligence Age
A framing of a period shaped by increasingly capable AI. In the book’s discussion of Sam Altman’s essay, it connects expanded intelligence to what people and institutions may be able to accomplish.
Intelligence gap
The separation between people who can use capable AI effectively and those who cannot. The book treats access, guidance, imagination, and agency as connected parts of that gap.
Iteration
Making an attempt, examining what happened, and revising the approach. A useful iteration responds to evidence or a clearer intention; producing another version alone does not establish improvement.
Large language model (LLM)
A model trained on large amounts of language and related data to work with sequences of tokens. It can support tasks such as writing, explanation, and coding, while still making factual and reasoning errors.
METR task horizon
A measure of task difficulty based on how long a human would take to complete tasks an AI succeeds at with a specified reliability. It is not the same as the time the AI ran or a guarantee of success on an unfamiliar job.
Multimodality
Working across more than one form of information, such as text, images, audio, or video. A model may handle several modalities directly; a connected workflow may combine multiple specialized tools.
Personal singularity
The book’s term for a shift in your sense of what you can make possible when connected AI capabilities bring previously impractical projects within reach. It describes a personal change in working life.
Personal superintelligence
A phrase used to imagine unusually capable AI assistance organized around an individual’s work and life. It does not, by itself, establish that a system meets a technical definition of superintelligence.
Prompt
The instructions, questions, and material given to an AI system. A useful prompt explains the purpose, supplies relevant context, describes the desired result, and gives a way to check whether it worked.
Prototype
An early version made to explore an idea or let someone try it. A prototype can be a page, sketch, recording, interaction, or functioning tool; choose the form that answers the question you need to investigate.
Recursive self-improvement
A proposed feedback process in which AI contributes to improving AI, and those improvements increase its capacity to contribute further. Whether this would accelerate rapidly depends on technical and practical conditions.
Singularity
A proposed period of technological change so profound that what follows becomes difficult to predict. Different writers use the term differently; Ray Kurzweil’s 2045 horizon is a forecast, not an established date.
Taste
The judgment to choose what belongs, what needs revision, and what is worth sharing. It develops through comparison, attention, practice, and learning how an intended audience experiences the work.
Token
A unit a model uses to represent information. In language tasks, a token may be a word, part of a word, or punctuation; token counts and word counts are not interchangeable.
Transformer
A neural-network architecture introduced in the 2017 paper “Attention Is All You Need.” Its attention mechanisms model relationships among parts of an input sequence; the architecture underlies many later language and multimodal systems.
Verification
Checking a claim or result against evidence outside the AI’s own reassurance. The appropriate check depends on the task: open a source, run a test, compare the output with the brief, or watch an intended user try it.
Vibe coding
Developing software by describing what you want to an AI coding tool and iterating on what it produces. A promising prototype still requires checks of functionality, usability, accessibility, and any relevant data handling.