APPLIED

workslop (noun, informal)

AI-generated work content (a report, email, or document) that looks legitimate but lacks the substance to actually move the task forward.

Example: "The draft looked polished, but it was workslop. The analyst spent longer figuring out what was wrong with it than she would have spent writing it herself."

AI slop (noun, informal)

Low-quality, generic AI-generated content produced in volume with little human direction or scrutiny.

Example: "The blog was publishing three AI-written articles a day, but readers quickly recognized it as AI slop and traffic dropped."

tool use (noun)

The ability of an AI system to call external functions, services, or software in order to do work beyond pure text generation. Tool use turns a model from a commentator into something that can retrieve data, take actions, or complete workflows.

Example: "Once tool use was enabled, the assistant could look up the shipment, update the CRM, and send the draft reply without leaving the chat."

Related: agent, integration, wrapper

reasoning model (noun)

A model optimized for multi-step thinking, structured problem solving, or harder analytical tasks. In practice, a reasoning model is chosen when accuracy and depth matter more than minimal cost or latency.

Example: "They used a reasoning model for the contract comparison because the job required careful step-by-step analysis rather than fast drafting."

Related: model, inference, confidence

prompt injection (noun)

An attack or failure mode where untrusted content manipulates an AI system by slipping in instructions that override or distort its intended behavior. Prompt injection is especially important when models read webpages, documents, or user-supplied text before acting.

Example: "The support ticket contained hidden text instructing the assistant to reveal internal pricing, a textbook prompt injection attempt."

Related: prompt, guardrails, responsible AI, tool use

orchestration (noun)

The coordination of models, tools, prompts, rules, and steps into a working AI process. Orchestration is what turns separate capabilities into a repeatable system.

Example: "The model generated the draft, another tool checked policy compliance, and a router assigned the result to the right queue; orchestration made the whole flow work."

Related: agent, agentic AI, hyperautomation

isolation (noun)

In AI systems, the condition where a model, assistant, or workflow is disconnected from the tools, data, or surrounding processes it would need to act usefully. Isolation often limits practical value even when the model itself is strong.

Example: "The assistant could explain the process well, but in isolation it could not access the inventory system or complete the task."

Related: integration, tool use, wrapper

human-in-the-loop (adjective)

Describing a workflow in which a person reviews, approves, corrects, or redirects AI outputs before the work is final. Human-in-the-loop design is often used where quality, safety, or accountability matters.

Example: "The system drafts the recommendation automatically, but a human-in-the-loop signs off before anything reaches the client."

hallucination (noun)

A confident but incorrect output produced by an AI system. A hallucination may sound plausible, but it contains fabricated facts, invented citations, or flawed reasoning.

Example: "The answer looked polished, but the case law it cited did not exist; that was a classic hallucination."

Related: confidence, AI literacy, guardrail

generative AI (noun)

AI that creates new content such as text, images, audio, code, or video based on patterns learned from training data. Its value lies in producing novel outputs rather than only classifying or retrieving information.

Example: "The team used generative AI to draft product descriptions, storyboard concepts, and first-pass campaign copy."

Related: generation, model, token

cost of intelligence (noun phrase)

The effective price of getting useful reasoning or output from an AI system. It includes not only raw model cost, but also the quality, speed, supervision, and operational overhead required to get dependable results.

Example: "A cheaper model looked attractive until the team counted the review time; the true cost of intelligence was higher than expected."

build vs. buy (noun phrase)

A decision about whether to create an AI capability internally or adopt an existing product, model, or platform. The trade-off usually involves speed, control, cost, risk, and long-term maintainability.

Example: "They chose buy over build for transcription, but built the internal workflow layer that made the transcripts useful."

AI labour (noun)

The practical work performed by AI systems in place of, or alongside, human effort. The term is often used when discussing what kinds of tasks AI can absorb, accelerate, or reshape economically.

Example: "Data cleanup used to take a coordinator a full day each week; now that AI labour handles the first pass, the team only reviews exceptions."

agentic AI (noun)

AI designed to operate with a degree of autonomy in pursuit of an objective. Agentic AI can plan, decide between actions, and use tools or workflows to complete multi-step work.

Example: "Instead of only suggesting next steps, the agentic AI opened the CRM, updated the record, and prepared the follow-up email."

Related: agent, automation, orchestration

agent (noun)

An AI system that can take actions toward a goal rather than only generating a single response. An agent typically reasons about the task, chooses tools or next steps, and adapts as new information arrives.

Example: "The support agent checked the order status, drafted a reply, and escalated the case when the refund rules were unclear."

wrapper (noun)

The layer of interface, instructions, and supporting logic built around a raw AI model to turn it into a usable product. The model itself typically just predicts likely next tokens. Everything that makes it feel like a coherent assistant (the chat window, the system prompt, safety checks, tool connections) is added around it.

Example: "Two companies used the exact same underlying model, but their wrappers made the products feel completely different to use."

Related: model, AI assistant

generation (noun)

The process by which an AI model produces new output based on patterns it learned during training. This might be text, an image, or another kind of material, depending on what the model is built to generate. For language models, it happens one token at a time, each one chosen based on everything that came before it.

Example: "Image generation took a few seconds longer than text generation because of how much more detail had to be produced."

Related: token, inference, model, generative AI

inference (noun)

The process of an AI model producing a response to a specific input, as distinct from the earlier process of training the model on data. Training happens once, or periodically, to build the model; inference happens every single time someone actually uses it to get an answer.

Example: "The company optimized inference speed so responses came back in under a second, without needing to retrain the model itself."

Related: model, token, reasoning model

confidence (noun)

A signal, sometimes shown to the user and sometimes just implicit, indicating how likely an AI model's output is to be correct.

Example: "The tool flagged its answer as low confidence, prompting the analyst to double-check the figure manually before using it."

AI economics (noun phrase)

(narrow, practical use) The practices and tradeoffs involved in managing the costs, investments, and returns of using AI within a business.
(broad, field of study) The broader notion of AI's economic impact on markets as a whole, an umbrella term covering ideas such as the AI economy, the cost of intelligence, and related concepts that have economic relevance.

Example: "Getting AI economics right meant weighing subscription costs against the hours of work actually saved."
Example: "Economists working in AI economics point to the widening gap between AI-native and legacy firms as evidence of a structural shift, not a temporary disruption."

AI economy (noun phrase)

The economic era in which AI is a first-tier, priced-in means of production, and acts as an economic actor in its own right.

*Example: "The AI economy meant a two-person startup could now compete with a service that used to require a fifty-person team."*

guardrail (noun)

Rules, filters, constraints, checks and balances, built into any part of an AI system and the environment in which the AI operates, with the purpose of keeping its behavior within prescribed and acceptable bounds.

Example: "A guardrail prevented the company's AI chatbot from ever quoting a specific price without a human sign-off."

bias (noun)

A systematic skew in an AI model's behaviour, traceable to an imbalance introduced at any point in the chain that produces and delivers it. Training data is one common source, but the same skew can enter through choices made in building or tuning the model, or through how the AI model is used.

*Example: "The team kept regenerating job-description drafts until one sounded like their existing hires — the bias wasn't in the model's training data, it was in which draft the team chose to keep."*

Related: responsible AI, guardrails, model

alignment (noun)

(outcome/state) How closely an AI system's behavior matches what the people using or building it actually intend, rather than what it was literally instructed to do. Because instructions are often incomplete or ambiguous, a well-aligned system fills those gaps the way a person would reasonably expect, while a poorly aligned one might technically satisfy the instruction while missing the intent behind it.
(technical, as a process) The training-time process of adjusting an AI model's behavior so that its outputs and actions match the intentions of the people training it. This is the work that produces the outcome described in sense 1, rather than the outcome itself.

Example: "The agent technically completed the task by deleting all the flagged emails, but that wasn't what anyone intended: a clear alignment failure."
Example: "The alignment of the new model was proving more challenging than they thought."

Related: guardrails, agentic, model

responsible AI (noun phrase)

The principles and practices an individual or organization applies to make sure its use of AI is safe, fair, compliant, and accountable. In practice, responsible AI usually covers topics like guardrails, bias checks, human review of high-stakes decisions, and clear ownership when something goes wrong.
*Example: "Responsible AI meant every high-stakes decision the system made still had a named person accountable for it."*

Related: guardrails, bias, alignment

AI assistant(noun)

An AI-powered product designed to help a person complete tasks through conversation or direct action. Built around one or more AI models, wrapped with an interface, instructions, and often tools that turn the raw model into something usable in a particular context. What a person calls "the AI" when using a tool like this is usually the assistant as a whole, not the bare model underneath it.
*Example: "I used an AI assistant to schedule and summarize my meetings."*

Synonym: copilot

Related: model, wrapper

hyperautomation (noun)

The combination of AI with existing automation tools to automate as much of an end-to-end business process as possible. Rather than automating a single isolated step, it strings multiple steps together into one continuous process, in many cases removing human-scale limitations with AI responsiveness.

Example: "Hyperautomation strung together document extraction, an AI drafting step, and the existing approval system into one unbroken process."

augmentation (noun)

The use of AI to support and enhance a person's existing work, without replacing their role in doing it. The person still makes the decisions and does the final work; the AI speeds up or improves a part of the process along the way.

Example: "The tool augmented the analyst's work by drafting a first pass, but the analyst still made every final call."

Related: automation, human-in-the-loop

AI literacy (noun phrase)

(individual)
A person's practical understanding of what AI can and can't do, sufficient to use it effectively and judge its output critically. It's less about technical knowledge and more about calibrated trust — knowing when to rely on an answer and when to check it.
*Example: "Basic AI literacy meant knowing to double-check a generated figure rather than assuming it was accurate simply because it sounded confident."*
(applied collectively)
The same capability, held or cultivated at the level of a team or organization rather than one person. Businesses tend to treat it as something with a floor: "AI literate" names the minimum level needed to work constructively with AI, and organizations invest in training and hiring specifically to clear it.
Example: "The organization lifted its AI literacy through training and hiring practices."

integration (noun)

(general use) The overall connectedness between AI and the environment it operates in: the processes, structures, and people around it, not just the technical systems it's wired into. A well-integrated AI shapes, and is shaped by, how a team actually works, rather than sitting apart from it as a separate tool people step outside their normal workflow to use.
(technical) The connections between an AI tool and an organization's existing systems and data, including how much latitude the AI tool is given to act on those systems directly. Without this kind of integration, using an AI tool's output usually means someone copying it manually into the system where it actually needs to live.

Example: "The AI was so well integrated with the team that their working practices elevated as a result."
Example: "Once the integration with the CRM was in place, the assistant could update customer records itself instead of someone copying its suggestions over manually."

Antonym: isolation

Related: tool use, agent, AI maturity

automation (noun)

The use of AI to carry out a task independently, without a person performing or directly overseeing each step. Unlike augmentation, the person's role moves from doing the work to setting it up and checking on it afterward, if at all.
*Example: "Automation meant the report was generated and sent every Monday morning with no one involved unless something went wrong."*

AI maturity (noun phrase)

(organizational journey) A measure of how deeply and effectively a person or organization has integrated AI into its actual working practices, as opposed to simply experimenting with it. It runs on a spectrum from isolated pilots and demos at one end to AI embedded as default practice at the other.
(skill/craft use) The level of skill someone brings to actually directing and working with AI, developed through practice rather than granted all at once. Immature use tends to produce generic, low-quality output (AI slop, or workslop in a workplace context) because the user isn't giving AI the direction, context, or scrutiny it needs. Mature use gets consistently strong results because the user has learned how to get them.

Example: "Running a chatbot demo didn't reflect AI maturity; having AI embedded in the actual approval workflow did."
Example: "Two people used the same tool on the same task. One got back something usable, the other got slop. The difference wasn't the tool; it was maturity."

Related: AI literacy, augmentation, integration, AI capability, AI slop, workslop

model (noun)

(technical) An AI system trained on large amounts of data to recognize patterns and generate output. It's the software that actually does the reading, reasoning, and writing whenever someone interacts with an AI tool. Named systems like GPT-4, Claude, and Gemini are all examples of models.
(general/practical use) Casual shorthand for an AI tool as a whole, standing in for the entire assistant experience rather than the underlying trained system in the strict technical sense. What a person actually interacts with is usually the model plus a wrapper of system instructions, tools, and interface built around it, but "the model" is often used loosely to mean all of it.

Example: "The company evaluated three different models before choosing the one that handled financial documents most reliably."
Example: "Just ask the model to summarize it" usually means "ask the AI assistant," even though what's really running is the model plus everything built around it."

tokenization (noun)

The process by which a body of text is broken down into tokens before an AI model can read or act on it. Tokenization happens automatically and invisibly, but understanding that it happens explains why AI systems sometimes handle language in ways that feel slightly literal or unpredictable. The model is working with fragments, not the whole picture at once.

Example: "Odd spacing or formatting in a document can affect how it's tokenized, and in turn how well the model understands it."

Related: token, token budget

tokenmaxxing (verb, informal)

Deliberately shaping a request, document, or workflow to extract the most value from a limited token budget. In practice, this often means packing in more substance or precision per token spent. Adapts the internet slang suffix "-maxxing" (pushing something to its fullest extent).

Example: "She rewrote the prompt three times, tokenmaxxing until every sentence pulled its weight."

token quality (noun)

(technical) A measure of how capable the tokens produced or consumed in an AI exchange are. It reflects the tier of model generating them: some services produce "premium" tokens with greater reasoning depth or accuracy per token, while lower-quality tokens are cheaper and faster. Each has its place depending on the task.
(general/evaluative use) Whether the tokens spent produced a result that is accurate, relevant, well-structured, and actually fit for its intended purpose. It's not just about producing an output, but the right output.

Example: "For a quick internal summary, lower-quality tokens do the job fine; for a client-facing legal analysis, we pay for premium tokens."
Example: "The summary was short and cheap to generate, but it missed the key risk: low token quality, even though token efficiency was high."

token efficiency (noun)

How much value or output is produced per token spent. Sometimes called the "how cheaply was this done" question. Low token efficiency, tokens spent without a commensurate return, is often described as waste.

Example: "The first draft of the prompt was bloated; the revised version got the same result in half the tokens, a big gain in token efficiency."

Related: token economics, token quality

token economics (noun phrase)

The practices and tradeoffs involved in managing tokens as a finite, costed resource in AI-driven work. This means deciding how many tokens to spend, how well they're spent, and what they produce. Token economics is an umbrella concept covering a growing family of related ideas, such as token budget, token efficiency, and token quality.

Example: "Good token economics isn't just cutting length; it's making every token count."

context window (noun)

The total token capacity a given AI model can hold in a single exchange. This is the hard technical ceiling set by the model itself, distinct from the token budget a person or organization chooses to actually spend within it. A model may be capable of far more than a given task uses.

Note: context window sits more naturally in the broader AI infrastructure glossary than in the token-economics family — it's included here because of its close working relationship with token budget, and will be cross-referenced rather than duplicated once that section exists.

Example: "The model's context window could hold the whole contract, but we only fed it the clauses relevant to the negotiation."

Related: token budget, token, model, context

token budgeting (activity, noun)

The act of setting, allocating, or managing a token budget, whether for a task, a person, or an organization.

Example: "Token budgeting is becoming as routine a management task as time or headcount budgeting."

token budget (noun)

(technical) The maximum number of tokens available for a given task or conversation with an AI system, covering both the instructions given to the model and the response it produces. A token budget is a chosen allocation. It may be set well below the model's maximum technical capacity (its context window) for reasons of cost, speed, or focus. Exceeding it forces a choice: trim the input, shorten the output, or split the work into stages.
(general/practical use) An amount of token spend allocated to a person, role, project, or piece of work. It's treated the way a team already treats a headcount or spending budget, and asking about it is increasingly routine for anyone starting a piece of AI-supported work.

Example: "We cut the appendix from the prompt; it was eating into our token budget."
Example: "Before starting the project, the new hire asked her manager: what's my token budget?"

token (noun)

(technical) The basic unit of text an AI model reads, processes, and generates. It typically takes the form of a word, a piece of a word, or a punctuation mark, rather than a whole sentence or paragraph.
(general use, by extension) A loose unit of measure for the amount of AI "work" something costs or takes. It's used casually, the way people say "gallons" of fuel, even by those who don't know the technical sense above.

Example: "The model splits sentences into tokens before generating a reply."
Example: "That report probably cost us a few thousand tokens to generate."