AI glossary

AI terms, in plain English

The words you’ll meet in AI tools and in conversations about AI at work, each with a definition, a comparison to something familiar and why it matters. Most are taught in The AI Stairway™ course, and each entry says where.

#

90-day test: A 90-day test, from The AI Stairway™ course, is a measured trial of one AI idea on one real process, run for about three months, with a baseline taken first, a measure agreed in advance and a stop rule. At the end, the numbers decide whether to scale it, change it and test again, or stop, rather than enthusiasm or a good demo.

A

AI anxiety: AI anxiety is an everyday name for worry about AI: about keeping up, making mistakes in front of others, or what it means for your job. It describes a common feeling, not a medical condition or a diagnosis. Mixed feelings about AI are common at work, including among people who use it every day.
AI champion: An AI champion is a colleague, usually not a specialist, who uses AI well in their own work and helps the people around them do the same: showing what works, answering everyday questions, and passing problems and ideas back to whoever sets the rules. The role usually sits alongside their day job.
AI fatigue: AI fatigue is the tiredness people feel from the amount of AI at work: new tools every month, constant announcements, output to check, and approval prompts clicked over and over. It describes a common experience, not a medical condition or a diagnosis, and it’s worth treating as a signal about workload and set-up, not a personal failing.
AI governance: AI governance is how an organisation decides and checks its use of AI: who may use which tools, with what information, who approves new uses, who owns each one, and how problems are reported and fixed. It covers the policy, the roles and the records, whether the organisation builds AI or only buys it.
AI literacy: AI literacy is the knowledge, skills and understanding people need to use AI sensibly at work: what the tools can and can’t do, what is safe to put into them, how to check what comes out, and where the opportunities and risks lie. The EU AI Act defines the term in Article 3(56).
AI maturity model: An AI maturity model is a staged description of how far an organisation has taken AI, from first experiments to AI built into how its work is designed. Each stage names the habits, rules and systems you would expect to see, so a team can tell where it stands and what comes next.
AI readiness: AI readiness is how prepared an organisation is to use AI well on real work: whether permission, access to the right data, governance and a named owner are in place, and whether people know how to use the tools safely. It is best judged for a specific piece of work rather than as one score for everything.
AI skill: An AI skill is a written-up way of doing one task, saved as instructions and sometimes example files, that an AI tool opens when it judges the task needs it. It lets everyone reuse a team’s best way of doing something, but it’s advice the AI usually follows, not a rule it must.
The AI Stairway™ to Success: The AI Stairway™ to Success is Scorchsoft’s own model of how AI becomes part of an organisation’s work, in seven steps from 0 to 6. It isn’t an industry standard. A person and their organisation can stand on different steps, and the step you’re on tells you what to learn or put in place next.

B

Baseline: A baseline is a measure of how a piece of work goes today, taken before you change anything: how long it takes, how often it goes wrong or what it costs. Without one, you can’t tell whether an AI change helped, and enthusiasm fills the gap. Measure the whole task, including checking time.

C

Carrying the outcome: Carrying the outcome is being the person who makes sure a piece of work achieves what it was for, not just that the task got done. In The AI Stairway™ course’s words: AI touches tasks; people carry outcomes. Whoever sends or signs off AI’s work still owns whether it’s right.
Connector: A connector is a link that lets an AI tool reach another app, such as your email, files, calendar or customer records, to read things and sometimes change them. Many connectors use a common standard called MCP (Model Context Protocol). What a connector can do depends on the permissions it’s given.
Context window: A context window is how much an AI model can hold on its desk at once: your messages, the conversation so far, instructions, files and anything a tool fetched, measured in tokens. It’s large in most tools as of 2026, but answers can get worse as it fills, and some tools quietly summarise older parts to make room.
Cost per task: Cost per task, from The AI Stairway™ course, is what it really costs to get one job finished to standard with AI: people’s time to prepare, check and fix it, plus the AI’s usage, including its working out, and a share of the licence. A cheaper model that needs more fixing can cost more per task than a dearer one.

D

Data protection impact assessment (DPIA): A data protection impact assessment (DPIA) is a documented check, made before you start, of how a use of personal data could harm people and how you will reduce that risk. UK GDPR Article 35 requires one where processing is likely to result in a high risk, and the ICO names AI among the technologies that can trigger one.

E

EU AI Act: EU AI Act is the usual name for Regulation (EU) 2024/1689, the European Union’s law on artificial intelligence. It sorts AI uses by risk: a few are banned, some need transparency, and high-risk uses such as some recruitment tools carry strict duties, which the Digital Omnibus of July 2026 put back to 2 December 2027.

G

Glue work: Glue work, as The AI Stairway™ course uses the phrase, is the re-typing, copying and pasting, chasing, remembering and reconciling a person does to move information between systems that don’t talk to each other. It’s often invisible and rarely in anyone’s job description, and it’s where the course looks for a first workflow to improve.

H

Human oversight: Human oversight is a person checking an AI system’s work in a way that could actually catch a problem, and stop or change the result, rather than clicking “Continue”. It needs the time, the knowledge and the authority to say no. The EU AI Act requires it for high-risk AI systems, in Article 14.

I

ISO/IEC 42001: ISO/IEC 42001 is the international standard for an artificial intelligence management system, first published in December 2023. It sets requirements for the policies, roles, risk assessments and reviews an organisation uses to govern the AI it builds or uses, and organisations can be certified against it by an independent body.

J

Jagged frontier: Jagged frontier is the name for the uneven edge of what AI can do: brilliant at some tasks, unreliable at others that look just as easy. It comes from a 2023 study of Boston Consulting Group consultants, led by Harvard Business School researchers, who did better with AI inside the frontier and worse just outside it.

L

Layers of control: Layers of control, from The AI Stairway™ course, are the four places a rule about AI can live, from weakest to strongest: a prompt asks, a skill shapes, a permission is enforced by software, and checks built into the process (code, or a person’s approval) make sure. A rule that must hold every time can’t live only in a prompt.
Lethal trifecta: Lethal trifecta is Simon Willison’s name for three things that, together, let an attacker steal data through an AI tool: access to private data, exposure to content from outsiders, and a way to send information out. Prompt injection can then turn the tool against you. Remove any one of the three and that route closes.

O

Organisation plan: An organisation plan is an AI account your employer sets up with an AI provider under business terms. On the terms we checked in October 2026, chats aren’t used to train its models by default, and your IT team can set controls, such as who has access and which features and connectors are on. It’s where work information belongs.

P

Personal account: A personal account is an AI account you set up for yourself with an AI provider, free or paid, under its consumer terms rather than a business agreement. Paying buys better models and higher limits, not business terms for your data, and, as of October 2026, your chats may be used for training, depending on a privacy setting you control.
Prompt injection: Prompt injection is an attack that plants hidden instructions for an AI inside something it reads, such as an email, a document or a web page, so that the AI follows the attacker instead of you. The UK’s National Cyber Security Centre says it may never be fully fixed, because today’s models don’t reliably separate instructions from data.

S

Shadow AI: Shadow AI is staff using AI tools at work that their organisation hasn’t approved or doesn’t know about, usually on personal accounts. It often comes from a gap in provision rather than bad intent: people find the tools useful and have no approved route. The main risk is work information leaving under personal terms.
Slop: Slop is work passed on without anyone thinking about it, most often AI’s: generic, padded or wrong, and it costs the reader time. It’s a behaviour, not a tool: the same AI can produce slop or careful work, depending on whether a person briefs it well and checks what comes back.
Solely automated decision: A solely automated decision is one taken with no meaningful human involvement, as UK GDPR Article 22A defines it. A significant one, with a legal or similarly significant effect on someone, must come with safeguards, including a way to contest it, and one using special category data is more tightly restricted (Article 22B).
Special category data: Special category data is personal data that UK GDPR Article 9 protects more strictly: racial or ethnic origin, political opinions, religious or philosophical beliefs, trade union membership, genetic data, biometric data used to identify someone, health, and sex life or sexual orientation. Using it needs a lawful basis and a separate condition.
Stakes ladder: A stakes ladder is a way to decide how hard to check what AI produces, by what’s at stake. The AI Stairway™ course uses four rungs: ideas, notes, work that goes out in your name, and anything touching care, money or rights, where you check every claim and figure against its source.
Stop rule: A stop rule is a rule, agreed before a trial starts, that says when to stop if it isn’t working: for example, if errors rise above a set level, or time saved stays below a target after a set number of weeks. Setting it in advance stops a failing pilot running on hope.

T

Three-second check: A three-second check is the habit, from The AI Stairway™ course, of asking three questions before you press Enter: Which account am I in? Would I be OK if this were read out? Can this AI act, or only answer? It takes moments, and it’s aimed at the mistakes that matter most.
Token: A token is a small chunk of text that an AI model reads and writes, often a word or part of a word. Models build every answer one token at a time, and AI providers count usage limits, speed and prices in tokens. Different tools split the same text into tokens differently.
Training on your data: Training on your data is when an AI provider uses your chats, files or feedback to help train its future models. As of October 2026, on personal plans it often depends on a privacy setting you control, while organisation plans generally don’t by default. The setting usually sits in the account’s privacy or data controls, and its name varies between tools.
Tramlines: Tramlines are The AI Stairway™ course’s way of sorting what goes into AI into three lanes: green (go, with nothing private in it), amber (with care, or with approval) and red (don’t). You decide the lane before you press Enter, and your organisation’s own rules always come first.
Trust check: A trust check is seven questions, from The AI Stairway™ course, to ask of any AI service before work information goes into it: who runs it, where your data is stored, whether it trains on your data, how long it’s kept, which terms apply, what it can reach and do, and whether your data protection lead would approve.
Trust ladder: A trust ladder is The AI Stairway™ course’s four levels of how far an organisation lets AI reach into its systems: inside one product, an approved AI tool, connectors that read your email and files, and connectors that act on them. Each rung up needs a clearer decision about who approves it, and for which data.

W

Workslop: Workslop is slop at work: AI-generated output that looks polished and finished but lacks the substance to move the task on, leaving the real thinking to whoever receives it. The name comes from researchers at BetterUp Labs (the research arm of BetterUp, a coaching company) and Stanford Social Media Lab, writing in Harvard Business Review in September 2025.