Explained in plain language

Glossary

The terms we underline with a dotted line in articles. Hover over an underlined word and the explanation appears right in the text — here they all are in one place.

agent: A program powered by AI that is given a goal and picks its own steps to reach it — reading data, calling tools and working for minutes without further input.
anonymisation: Removing or replacing details that could identify a specific person. Done before sending data to an AI or any other outside service.
API: An interface that lets programs talk to each other without a human. Through it your app can request data or a function from another service.
API key: A secret string your app uses to sign in to another service API. Anyone who gets the key can use the service on your account, so it is never shared.
artifact: A finished output the AI produces as a standalone document or app — a table, a web page or a calculator you can open and share right away.
ATS: Software companies use to receive and sort job applications. Your CV usually passes through it before any human reads it.
BATNA: The best option you have if the negotiation fails. The stronger your BATNA, the calmer you can negotiate.
cash flow: The real movement of money in and out of your account. A business can be profitable and still fail when it runs out of cash.
cloud: Someone else's servers where a service runs and your data lives. You reach them over the internet and the provider handles the hardware.
connector: A link between the AI and a specific service such as Gmail, Calendar or Drive. Once connected, the AI reads and writes there directly, with no copy and paste.
context window: How much text the model can hold in mind at once, the prompt and its own answer together. Go past the limit and the start of the conversation drops out.
CRM: A system where a company tracks customers and deals — contacts, past conversations and the status of each opportunity.
CSV: A plain text table format where values are separated by commas or semicolons. Excel and Google Sheets open it, and nearly every program reads it.
decision fatigue: The drop in decision quality after making many choices in a day. It is why important calls belong in the morning and routine ones get automated.
deep research: A mode where the AI spends minutes browsing dozens of sources on its own and returns a summary with links. Good for market research or vendor comparisons.
DPA (data processing agreement): A data processing agreement between you and a vendor that handles personal data. GDPR requires one before you put company data into an outside tool.
endpoint: A specific API address you send a request to. Each endpoint does one thing, such as returning a list of invoices or creating a contact.
ERP: The backbone company system holding inventory, accounting, orders and payroll in one place. Data entered in one part flows into the rest.
EU AI Act: The European regulation that sorts AI systems by risk level and sets duties accordingly. The strictest rules apply to high risk uses such as hiring or lending.
few-shot (examples in the prompt): A prompting technique where you add a few finished examples of the output you want. The AI picks up format and tone from them better than from a description.
GDPR: The European data protection regulation. It sets when you may collect data about people, how long you may keep it and what rights you must honour.
GTD: The Getting Things Done method: move every task out of your head into one list, sort tasks by context and review the system regularly.
hallucination: When an AI makes an answer up and states it confidently as fact. It happens most with numbers, quotes and links, so those need checking.
human-in-the-loop: Keeping a person in the deciding step of an automation — the AI prepares the output, but a human approves or sends it.
Inbox Zero: The habit of emptying your inbox to zero — each email is deleted, answered, delegated or turned into a task.
language model (LLM): A program trained on huge amounts of text that predicts what words come next. ChatGPT, Claude and Gemini are all built on this.
markdown: A way of formatting text with plain characters — asterisks make bold, hashes make headings. It stays readable for both people and machines.
MCP: An open standard for connecting AI to tools and data. One integration is enough, and it works across different applications.
OCR: Turning an image or a scan into text you can search and copy. Used for photographed receipts, contracts and old documents.
open source: Software whose source code is public, so anyone can inspect it, change it and run it themselves. Usually free of charge, but maintenance costs you time.
orchestration: Coordinating several steps or several AI agents so they follow on from each other. Something has to set the order, pass results along and catch errors.
pivot table: A tool in Excel and Google Sheets that turns a list of rows into a summary — sales by month and product, for instance, with no formulas.
Pomodoro: Working in blocks of about 25 minutes with a short break between them. After four blocks you take a longer break.
prompt: The instruction you give an AI — the text that tells it what to do. The more specific the prompt, the better the result.
RSS: A format websites use to publish their list of new posts. In a reader you follow dozens of sources in one place, with no algorithm in between.
scheduled task: An instruction that runs by itself at a set time or interval — preparing a weekly summary every Monday morning and emailing it, for example.
second brain: An outside notes system where you store ideas, tasks and links so you do not have to remember them. Your head is then free to think, not to archive.
shadow AI: Employees using AI tools without the company knowing. The main risk is company data ending up in services nobody has vetted.
SSO: One login for all company applications. Staff remember no extra passwords, and when someone leaves you disable a single account.
subagent: A helper agent the main agent hands a subtask to. It works in its own context and returns only the result, keeping the main thread clean.
system prompt: A standing instruction the AI receives before every conversation, setting its role, tone and limits. Users normally never see it.
text expander: A tool that turns a short abbreviation into a longer block of text. Type two letters and a full reply or your billing details appear.
time-blocking: Planning work straight into the calendar — every task gets a specific slot. A wish list turns into a schedule with visible limits.
token: The chunk of text a model breaks language into — roughly a word or part of one. Input length and pricing are both measured in tokens.
TTS (text-to-speech): Turning written text into spoken audio. Useful when you would rather listen to a long document on the move than read it.
webhook: An automatic message one service sends another the moment something happens. Instead of you polling it, the service calls you.