Glossary
Every new term introduced in the lessons, in one place. Each entry links to the lesson where it appears.
- .env file
- A text file in the project folder where the secrets are kept, one per line, in the format NAME=value. It hides nothing from the agent, which can read it, nor from the browser, if the key ends up in a web page. (Secrets, permissions and spending caps)
- .gitignore
- A file with the list of what Git should leave out when you run git add. It stops a new file from going into a commit by accident, but it does not remove from Git what was already tracked, and it does not stop anyone, the agent included, from reading it. (Secrets, permissions and spending caps)
- 3-D Secure
- An extra security step in a card payment, where the customer's bank asks for a code or a confirmation in its app. It protects the customer and it protects you. (Taking payments online)
- Action (in an automation)
- Each step the automation takes after the trigger: write in a spreadsheet, ask the AI for a text, send a message. (Serious automation: Zapier, Make and n8n)
- ADR (alternative dispute resolution, RAL)
- Bodies outside the courts that a consumer can turn to for settling a consumer dispute. The site should say which ones. (Legal obligations in Portugal)
- Adversarial reviewer
- A reviewer who is asked to look for what can go wrong, instead of confirming that it is fine. (Appendix: cross-auditing between two model families)
- Agent SDK
- A library that puts the Claude Code engine inside a program of your own: the same loop of reading, using tools and answering, but controlled by code. (Agent SDK: building your own agent)
- AGENTS.md
- A text file with instructions that Codex reads at the start of every session. Other agents recognise it too. (CLAUDE.md and AGENTS.md: giving the agent context)
- AI Act
- Regulation (EU) 2024/1689, the European law on artificial intelligence. It came into force in 2024 and applies in phases. (What never to paste: data, rights and the law in 5 minutes)
- AI agent
- An AI to which you give a goal and which decides by itself the steps to get there: search, open pages, fill in, correct its path. In a conversation, you are the one who decides the next step. (What comes next: agents for non-programmers)
- AI literacy
- Knowing enough about AI to use it usefully and carefully: what it does, where it goes wrong and what risks it carries. (What never to paste: data, rights and the law in 5 minutes)
- AI use policy
- A page written by the company or organisation that says which AI tools are used, for what, which data never goes in and who answers for the result. (AI in a team or company: rules, costs and literacy)
- Alias
- An extra address that delivers to the same mailbox. For example, invoices@ and hello@ can both reach the same person, without paying for a separate mailbox for each. (Professional email on your own domain)
- API
- A program that answers requests from other programs, for example “give me the list of products”, instead of showing a page to a person. (Project: an API with database, tests and deploy)
- API key
- A secret code that links a program to your pay-per-use account. Whoever has it spends in your name, so you keep it like a password. (Install and sign in: Claude Code and Codex on Windows, Mac and Linux)
- Artificial intelligence (AI)
- Computer programs that do tasks that used to seem to need a person, such as recognising faces, translating or writing text. (What AI is (and what it is not))
- ATCUD
- A unique code that must appear on invoices issued by software certified by the Tax Authority, together with a QR code. (Taking payments online)
- Attachment
- A file you add to the conversation (a PDF, a spreadsheet, a photograph) for the AI to work on. Attaching is sending: the same rules apply as for pasting text. (Working with documents: PDFs, spreadsheets and images)
- Authentication
- Proving to the program who you are, by signing in with your account in the browser or with an API key. (Install and sign in: Claude Code and Codex on Windows, Mac and Linux)
- Auto-renewal
- An option that renews the domain by itself before it expires, charging the card on file. It stops you losing the name by forgetting. (The domain: choosing, registering and never losing it)
- Automatic confirmation
- A message (email or text) the customer receives right after booking, with the day, time, place and how to cancel. It proves the booking exists, for them and for you. (Online bookings and appointments)
- Automatic memory
- Notes that the agent writes by itself to remember preferences and corrections in later sessions. (Context, compaction and memory)
- Automatic test
- A small function that calls your code with an example whose answer is known, and warns you if the result is different. (Asking an agent well: small tasks, done criteria and what a test is)
- Away message
- An automatic reply that WhatsApp Business sends outside your opening hours, saying when you will answer again. (Social media and WhatsApp, with a plan)
- Backup
- A copy of the site and the contacts kept outside the site itself. If something breaks or disappears, you restore from it. (Keeping it alive and safe)
- Benchmark
- A public test, the same for everyone, used to compare AI models: exams, maths problems, programming. It measures those tasks, not necessarily your work. (Judging quality: blind tests and rubrics)
- Bias
- A systematic tendency of a result towards one side, inherited from the data the AI learned from. Example: showing almost only men when you ask for a portrait of someone who works in medicine. (Ethics and law: the AI Act, GDPR and copyright)
- Bio
- The short text at the top of your profile on a social network. It is the place to say what you do, where you are and put the link to your website or bookings. (Social media and WhatsApp, with a plan)
- Blind test
- Comparing answers without knowing which tool or which request they come from, so that opinion about the brand does not influence the score. (Judging quality: blind tests and rubrics)
- Branch
- A parallel line of snapshots in Git, where you can experiment without touching the main branch. (Subagents, models and costs)
- Broadcast list
- A list of contacts you send the same message to; each person receives it as a private chat and does not see the others. It should contain only people who asked to receive news. (Social media and WhatsApp, with a plan)
- Built-in AI
- AI features that already come inside programs you use (email, Word, phone), with nothing extra to install. (The tools: which to use for what)
- Business listing
- Your business's page on Google and on maps, with name, address, hours, phone, photos and reviews. For many people it is the first thing they see when they look for a place nearby. (Online presence: what it is and what you need)
- Business listing (Google Business Profile)
- The free page Google shows when someone searches for your business, with opening hours, phone, photos, reviews and your spot on the map. You manage what is on it once you prove the business is yours. (Showing up on Google and on maps)
- Catalogue (WhatsApp Business)
- A shop window inside WhatsApp Business where you put photos, names and descriptions of your products or services, so the customer can see them without leaving the conversation. (Social media and WhatsApp, with a plan)
- Certificate (HTTPS)
- A digital document proving the site is what it says it is and encrypting the connection. It has an expiry date; if it lapses, the browser shows visitors a red warning. (Keeping it alive and safe)
- Channel manager
- A tool that links an accommodation's calendars to the various booking platforms and keeps them identical, so a booking in one place blocks the dates in the others. (Online bookings and appointments)
- Chat assistant (chatbot)
- A program you talk to in writing or by voice, such as ChatGPT, Claude or Gemini. It answers each message you send. (Your first conversation: asking well)
- CI (continuous integration)
- The practice of running the tests automatically, on a separate machine, every time someone sends a change. (CI/CD with agents: GitHub Actions and PR review)
- Citation (in an AI answer)
- A link or a number next to a sentence, showing where the AI says it took the information from. You have to open it to confirm: the page may exist and not say that. (Researching with AI without being fooled: sources and citations)
- CLAUDE.md
- A text file with instructions that Claude Code reads at the start of every session. (CLAUDE.md and AGENTS.md: giving the agent context)
- Cloned voice
- An imitation of a person's voice made with AI from recordings of them, for example from videos posted on social media. (AI scams: cloned voices, fake videos and phishing)
- CNPD
- Comissão Nacional de Protecção de Dados (the National Data Protection Commission): the authority that supervises the GDPR in Portugal and receives complaints. Website: cnpd.pt. That is if you are in Portugal; elsewhere use your country's equivalent. (Ethics and law: the AI Act, GDPR and copyright)
- Code word
- A word agreed within the family, spoken aloud only, to confirm who is on the other end of an emergency call or message. (AI scams: cloned voices, fake videos and phishing)
- Coding agent
- An AI program that works in a folder on your computer: it reads files, makes a plan, changes files and runs commands. Depending on the mode, it asks your permission for actions with consequences. (What a coding agent is (and why it is not a chat))
- Command
- A short instruction typed in the terminal, such as pwd or mkdir. Typed by hand, it only runs when you press Enter; if you paste several lines at once, some may run straight away. (The terminal without fear)
- Command (slash command)
- An instruction that starts with / and that the tool recognises, like /model or /diff. Type / to see the list. (Commands, skills and repeatable tasks)
- Comment
- A note in the code for people to read. The computer ignores it. It starts with # in Python and with // in JavaScript. (Reading code, the minimum: variables, functions and errors)
- Commit
- A snapshot of the folder at a given moment, with one sentence saying what changed. It is a point you can go back to. (Essential Git with the agent at your side)
- Compaction
- Swapping the conversation for a summary of it, to free up space in the context window without starting from zero. (Context, compaction and memory)
- Company account
- An account with an AI service that the company signs up for and pays, with an administrator who gives and removes access and chooses the privacy settings for everyone. (AI in a team or company: rules, costs and literacy)
- Connection tool
- A program that links other programs together (email, spreadsheet, AI) with rules of the type "when this happens, do that". Zapier, Make and n8n are examples. (Automating your first workflow)
- Context
- Everything the agent has in mind during a conversation: your requests, its answers and the files it has read. A new conversation starts with a clean context. (When it goes wrong: undo, restart and understand the error)
- Context window
- Everything the AI can "see" at once in a conversation. When the conversation gets too long, what is at the start falls out of the window and the AI stops taking it into account. (Why it forgets and why it makes things up)
- Cron
- A Linux and macOS service that runs scheduled commands, configured with the crontab command. (Final project: a system that works alone, with brakes)
- Cross-auditing
- Asking a model from another family to review an agent's plan or code, before a person decides. (Appendix: cross-auditing between two model families)
- Current folder
- The folder the terminal is in at that moment. Commands act there unless you say otherwise. The pwd command shows which one it is. (The terminal without fear)
- Custom instructions
- Rules you write once in the settings that apply to every conversation, for example "answer in plain English, short". Each service gives them a different name. (Projects, custom instructions and memory)
- Cut-off date
- The date up to which the texts the model was trained on go. What happened after that, it does not know, unless search is switched on. (Why it forgets and why it makes things up)
- Deep research
- A mode in some assistants in which the AI spends several minutes reading many pages and hands over a report with sources. More pages do not mean more truth. (Researching with AI without being fooled: sources and citations)
- Deepfake
- A fake video, image or sound made with AI, which makes a real person appear to say or do what they never said or did. (AI scams: cloned voices, fake videos and phishing)
- Diff
- The list of differences between two versions of a file: lines marked − come out, lines marked + go in, lines with no sign stay the same. (Reading code, the minimum: variables, functions and errors)
- DKIM
- A digital signature your provider adds to every email you send. The recipient checks it with a public key stored in your domain's DNS, and so knows the message was not altered on the way. (Professional email on your own domain)
- DMARC
- A record in the DNS that tells the recipient what to do with emails that fail SPF and DKIM (let them through, put them in spam or reject them) and where to send the reports. (Professional email on your own domain)
- DNS
- The system that turns the domain name into the right places: where the website is and where the email goes. It works like a list of road signs. (The domain: choosing, registering and never losing it)
- Domain
- The name of your address on the internet, for example padariaamaral.pt. It is the address: it says where to find you, but it is neither the website nor the email. (Online presence: what it is and what you need)
- Done criterion
- The sentence that says how you check that a task is finished, for example “when python3 -m pytest passes”. (Asking an agent well: small tasks, done criteria and what a test is)
- dontAsk mode
- A permission mode in which the agent never asks: it does what is pre-approved and refuses everything that would need authorisation. (Agent SDK: building your own agent)
- Electronic Complaints Book (Livro de Reclamações Electrónico)
- The official platform (www.livroreclamacoes.pt) where a consumer files a complaint. Anyone who sells or provides services to the public, even online only, must offer it on the site with a visible link. (Legal obligations in Portugal)
- Embedding
- The way the AI turns a sentence into a point on a "map" of meanings. Sentences with a similar meaning sit close to each other, even with different words. (How it works inside: training, embeddings and RAG)
- Endpoint
- An address of the API that does one thing, such as GET /produtos to list or POST /produtos to create. (Project: an API with database, tests and deploy)
- Ephemeral disk
- A hosting service's disk that is rebuilt on every restart or new deploy, losing the files created in the meantime. (Project: an API with database, tests and deploy)
- Examples in the prompt (few-shot)
- Showing the AI 1 or 2 examples of the result you want, so that it follows the same style and shape. This is called few-shot prompting, meaning "with a few examples". (Advanced prompting: roles, steps, examples and iterating)
- Execution log
- The list that the automation tool keeps of every time a flow ran: what went in, what came out of each step and where it failed. (Serious automation: Zapier, Make and n8n)
- Exit code
- A number that a program returns when it ends. 0 means “went well”. In the hooks of these two tools, 2 means “block”. (Hooks: automatic checks the agent cannot skip)
- Export (a website)
- Taking a copy of the site and its contents so you can move them somewhere else. Not every service lets you; it is worth knowing before you choose. (The website: what it needs and how to put it up without coding)
- Extension
- The letters after the last dot in a file name (.pdf, .html, .py). They tell the computer what kind of file it is. (Folders, files and an editor (Windows, Mac and Linux))
- Fee
- What the payment provider keeps from each sale, usually a small percentage plus a fixed amount. Reference values are on the Current state page. (Taking payments online)
- Fine-tuning
- Extra training, after the main training, with examples of good answers and with people choosing the better of two. It is what turns a text guesser into an assistant that holds a conversation. (How it works inside: training, embeddings and RAG)
- Formula (in a spreadsheet)
- An instruction you write in a cell so that the spreadsheet does the sum by itself, for example =SUM(B2:B20). Unlike the AI, the spreadsheet does not get sums wrong. (Working with documents: PDFs, spreadsheets and images)
- Function
- A named block of code, like a recipe: it takes ingredients, always does the same steps and gives back a result. Using it is called "calling" the function. (Reading code, the minimum: variables, functions and errors)
- GDPR
- General Data Protection Regulation. The European law that protects personal data, applied since May 2018. (What never to paste: data, rights and the law in 5 minutes)
- Generated image
- An image created by an AI from a written request, with no camera. It can look like a real photograph. (Creating with AI: images, voice, video and slides)
- Generative AI
- AI that creates new things (text, images, sound) from patterns it has learned from many examples. (What AI is (and what it is not))
- Git
- A program that keeps snapshots (versions) of a folder and lets you go back to any of them. (Essential Git with the agent at your side)
- git restore
- A Git command that puts a file back from a saved version. On its own it takes it from the staging area (what you marked with git add); with --source=HEAD, from the last commit. It wipes the unsaved changes in that file. (When it goes wrong: undo, restart and understand the error)
- GitHub
- An Internet service where you keep a copy of your repositories. It also publishes websites, with GitHub Pages. (Essential Git with the agent at your side)
- GitHub Pages
- A GitHub service that publishes on the Internet the HTML pages kept in a repository. (Project: a personal site, from zero to online)
- Hallucination
- A made-up answer, said with confidence, that sounds good but is not true. It happens because the AI writes what seems likely. (Why it forgets and why it makes things up)
- High-risk AI system
- In the AI Act, an AI system used in areas that can change someone's life (employment, education, credit, essential public services, justice, among others). Whoever makes it and whoever uses it have stronger obligations. (Ethics and law: the AI Act, GDPR and copyright)
- Hook
- A command that the tool runs by itself at a fixed moment, for example before every edit or when the agent finishes answering. (Hooks: automatic checks the agent cannot skip)
- Hosting
- The service where your website's files are stored and connected to the internet, so anyone can open them. It is the house where the website lives. (Online presence: what it is and what you need)
- HTML
- The language of web pages. A .html file opens in the browser and shows the page. (Your first task: plan, diff and permissions)
- HTTPS
- The protected version of a website address. It shows as a padlock in the browser and means the connection between the visitor and the site is encrypted. Today it is a must in practice: without it, the browser shows a warning. (The website: what it needs and how to put it up without coding)
- Human review
- Reading, checking and correcting what the AI did before using or sending it. The last word is always yours. (Everyday exercises)
- iCal
- A common calendar format, a link that other programs can read. It is used to share the busy dates between your calendar and the platforms. (Online bookings and appointments)
- Independent sources
- Sources that each reached the information in their own way. Two news items that copy the same press release, or two pages from the same website, count as one. (Researching with AI without being fooled: sources and citations)
- Iterate
- To improve an answer little by little, with short follow-ups in the same conversation: "shorter", "more formal", "remove the last paragraph". (Your first conversation: asking well)
- Knowledge base
- A set of documents that you give to an AI tool for it to consult before answering: regulations, manuals, frequently asked questions. (How it works inside: training, embeddings and RAG)
- Label (in place of personal data)
- A mark you put in place of a piece of personal data before pasting a text, for example "[NIF]" or "customer A". Swapping data for labels is pseudonymising: the text is still not anonymous, because the context may give the person away (lesson 5). (Everyday exercises)
- Labelling of generated content
- Stating that an image, a sound or a video was created or altered with AI, for example with the sentence "Image created with AI". For deepfakes, European law requires it. (Creating with AI: images, voice, video and slides)
- Language model
- A program trained on an enormous amount of text to guess the next word. It is the engine inside assistants such as ChatGPT, Claude or Gemini. (What AI is (and what it is not))
- Legal basis (GDPR)
- The reason, set out in Article 6 of the GDPR, that allows you to process personal data: for example, the person's consent, a contract with them or a legal obligation. (Ethics and law: the AI Act, GDPR and copyright)
- Manual mode
- In Claude Code, the mode in which the agent asks permission before editing files and before most commands. It is the right one for learning. (Your first task: plan, diff and permissions)
- MCP (Model Context Protocol)
- An open standard that defines how an agent connects to external tools, such as a calendar, a database or a browser. (MCP: connecting external tools (and prompt injection))
- MCP server
- A small program that offers tools to an agent through MCP, for example “tell the time in a time zone”. (MCP: connecting external tools (and prompt injection))
- Memory (between conversations)
- Notes the service keeps about you from your conversations, to use later. You can see them, delete them or switch the feature off in the settings. (Projects, custom instructions and memory)
- Merge
- Bringing the snapshots of one branch into another, usually from an experiment branch into the main one. (Subagents, models and costs)
- Model family
- Models from the same company, trained in a similar way. They tend to share habits and blind spots. (Appendix: cross-auditing between two model families)
- Official app
- The app published by the company that makes the service itself. There are fake copies in the stores with similar names. (The tools: which to use for what)
- Official source
- The body that decides or publishes the information: the service's own website, the notice of the application, the Diário da República (Portugal's official gazette), the medicine leaflet. It is the one that rules when there is doubt. (Researching with AI without being fooled: sources and citations)
- Online presence
- The set of places on the internet where a customer finds you and contacts you: website, Google listing, email, social media and WhatsApp. It is not just having a web page. (Online presence: what it is and what you need)
- Orchestration
- Deciding which agents run, in what order, with what data, and who merges the results. (Multi-agent and orchestration)
- Parallelising
- Making several independent parts of the work run at the same time, each in its own agent. (Multi-agent and orchestration)
- Password manager
- An app that keeps a different password for each service, protected by a single master password, and fills it in for you. There are free and paid options. (Keeping it alive and safe)
- Path
- The address of a file: the list of folders to reach it, separated by \ on Windows and by / on Mac and Linux. (Folders, files and an editor (Windows, Mac and Linux))
- Pay per use
- Paying according to the amount of work the AI does, instead of a fixed amount per person per month. It is the most common way when AI is connected to automations. (AI in a team or company: rules, costs and literacy)
- Payment link
- A link you send by WhatsApp, email or message that takes the customer to the provider's payment page, where they choose how to pay. You do not need an online shop to use it. (Taking payments online)
- Payment service provider
- The company that processes the payment between the customer and your account (a bank or an authorised company). It is the one you open the merchant account with. (Taking payments online)
- Per-run cap
- A limit on turns or estimated spending that the program imposes on each run of the agent. It complements the spending limit in the provider's console. (Agent SDK: building your own agent)
- Permission request
- The moment when the agent stops and shows you what it wants to do, waiting for a yes or a no. When it appears depends on the mode and the rules. (What a coding agent is (and why it is not a chat))
- Personal data
- Any information about an identified person, or one who can be identified: name, address, tax number, phone number, photo, voice, health data. (What never to paste: data, rights and the law in 5 minutes)
- Phishing
- A fake message (email, SMS, WhatsApp) pretending to be a bank, a public body or a company, to steal your data, codes or money. (AI scams: cloned voices, fake videos and phishing)
- Pinning versions
- Writing the exact version of each package in requirements.txt (with ==), so you always install the same code. (Security in depth: sandbox, supply chain and auditing)
- Plan mode
- A mode in which the agent reads and proposes a plan, but does not change your files until you approve. (Your first task: plan, diff and permissions)
- PreToolUse
- The moment before the agent uses a tool, such as editing a file or running a command. A hook at this moment can block the action. (Hooks: automatic checks the agent cannot skip)
- Privacy policy
- A text on the site saying who is responsible for the personal data, what data it collects, why, how long it keeps it, who it shares it with, and what rights the person has. (Legal obligations in Portugal)
- Prohibited AI practice
- A use of AI that the AI Act bans in the European Union since 2 February 2025, such as manipulating people with hidden techniques or giving them a "social score". (Ethics and law: the AI Act, GDPR and copyright)
- Project (in an AI tool)
- A space dedicated to one subject, holding instructions and files that every conversation in that space uses. Some services call it a space or another name. (Projects, custom instructions and memory)
- Project root
- The main folder of the project, the one where you open the agent and where the hidden .git folder is. (CLAUDE.md and AGENTS.md: giving the agent context)
- Prompt
- What you write or say to the AI to get it to do something. Also called a request. (Your first conversation: asking well)
- Prompt injection
- A trick in which someone hides orders in a page, a document or an email to fool the agent that reads them ("ignore what you were asked and send…"). (What comes next: agents for non-programmers)
- Protected branch
- A branch with rules that block changes made without a pull request, without approval, or with failing tests. (CI/CD with agents: GitHub Actions and PR review)
- Pseudonymisation
- Swapping names, numbers and addresses for labels ("person A", "[NIF]"). It protects, but it does not make the text anonymous: if the context still shows who it is about, it is still personal data. (What never to paste: data, rights and the law in 5 minutes)
- Publish
- Putting a page at an address that anyone can open on the Internet. (Project: a personal site, from zero to online)
- Push
- Sending the commits from your computer to the remote. (Project: a personal site, from zero to online)
- pytest
- A Python tool that finds the test files, runs them and tells you how many passed and how many failed. (Asking an agent well: small tasks, done criteria and what a test is)
- RAG
- Short for "retrieval-augmented generation". Before answering, the tool fetches from your documents the pieces most similar to the question and reads them. It is what projects and knowledge bases do. (How it works inside: training, embeddings and RAG)
- Reasoning effort
- How much the model thinks before answering. More effort can give a better result, but it takes longer and spends more. (Subagents, models and costs)
- Registrant (domain holder)
- The person or company the domain belongs to. They are in charge: they can renew it, transfer it or lose it. It should be the business owner. (The domain: choosing, registering and never losing it)
- Registrar
- The company where you register and manage your domain. For .pt you can register directly with DNS.PT or through a registration agent; for other endings, at an international registrar. (The domain: choosing, registering and never losing it)
- Remote
- A copy of the repository kept somewhere else, such as GitHub. The usual name for the main remote is origin. (Project: a personal site, from zero to online)
- Repeatable task
- A task that, with the same request, gives the same result every time you do it. (Commands, skills and repeatable tasks)
- Repository
- A folder controlled by Git, with the whole history kept in a hidden subfolder called .git. (Essential Git with the agent at your side)
- Repository secret
- A value stored in GitHub's settings, such as an API key, that workflows use by name without showing it. (CI/CD with agents: GitHub Actions and PR review)
- Restore point
- A snapshot of the state of your files at a certain moment, to which you can go back. In Claude Code they are called checkpoints, and one is created with each request you make. (When it goes wrong: undo, restart and understand the error)
- Revoke
- To cancel a key in the provider's console. From that moment it stops working, for you and for anyone who has it. (Secrets, permissions and spending caps)
- Right of withdrawal (direito de livre resolução)
- The consumer's right to cancel a distance purchase within 14 days, with no reason given. The law sets out exceptions. (Legal obligations in Portugal)
- Role (in a prompt)
- An instruction about who the AI should pretend to be and who it is talking to, for example "you are an accountant explaining to a shopkeeper". It changes the tone and level of the answer; it does not give the AI real knowledge. (Advanced prompting: roles, steps, examples and iterating)
- Rubric
- A short list of criteria, each with a scale of scores (for example, from 1 to 5) and a description of what each score is worth. It is used to assess different answers in the same way. (Judging quality: blind tests and rubrics)
- Sandbox
- A box around the agent's commands that limits where they can write, what they can read and whether they use the network, even when nobody asks anything. (Security in depth: sandbox, supply chain and auditing)
- Scanned PDF
- A PDF made from a photograph or a scanner. It looks like text, but to the computer it is an image: the AI has to "read" it and makes more mistakes, especially with numbers. (Working with documents: PDFs, spreadsheets and images)
- Scheduled task
- A request you leave set up in an AI assistant for it to do by itself at a certain time, once or repeatedly. The result appears in the conversation or as a notification. (Automating your first workflow)
- Scheduling
- An instruction to the operating system to run a program at a set time, for example every day at 7:00. (Final project: a system that works alone, with brakes)
- Search Console
- A free Google tool where you prove the website is yours and see which searches bring people to it and whether there are problems. (Showing up on Google and on maps)
- Search switched on
- When the assistant goes onto the internet to look for current information and shows the sources it took it from. Without it, it answers only with what it learned in training. (The tools: which to use for what)
- Secret
- Any piece of information that gives access to an account or a service: a password, an API key, a token. (Secrets, permissions and spending caps)
- Self-critique (by the AI)
- Asking the AI to reread its own answer, point out the problems and fix them. It helps with clarity and form; it does not replace checking the facts. (Advanced prompting: roles, steps, examples and iterating)
- Self-hosting
- Running a tool on a computer of your own (or of your IT supplier), instead of using it on the servers of the company that makes it. (Serious automation: Zapier, Make and n8n)
- SEO
- Everything that helps Google understand what your website says and show it to people who are searching for that. Short for "search engine optimisation". (Showing up on Google and on maps)
- Single calendar
- One calendar where every booking or appointment goes in, wherever it comes from (phone, WhatsApp, website, platforms). It is what stops two people getting the same slot. (Online bookings and appointments)
- Sitemap
- A file with the list of your website's pages, so Google finds them faster. Most website builders create it automatically. (Showing up on Google and on maps)
- Skill
- A folder with a SKILL.md file that holds the instructions for a task. You call it by name, or the agent uses it when its description fits your request. (Commands, skills and repeatable tasks)
- Spending cap
- A limit you turn on in the provider's console that blocks requests when spending reaches an amount. An alert is something else: it only warns you. Enforcement may not be instant, so leave some margin. (Install and sign in: Claude Code and Codex on Windows, Mac and Linux)
- SPF
- A record in the DNS that lists the servers allowed to send email on behalf of your domain. Without it, the recipient may suspect the message is fake. (Professional email on your own domain)
- SQLite
- A database that keeps everything in a single file and comes included with Python, with no separate server. (Project: an API with database, tests and deploy)
- State file
- A file in the project where what is done, what is left and the decisions taken are written down, so you can resume the work in another session. (Context, compaction and memory)
- Stop button
- A simple, agreed way of halting an automatic system. Here, a file called PARAR: if it exists, the agent does not run. (Final project: a system that works alone, with brakes)
- Subagent
- A second agent that the main one launches for a separate task. It works with its own context window and returns only a summary. (Subagents, models and costs)
- Supply chain (dependency chain)
- All the packages your project depends on, and the packages those depend on. Each one is someone else's code running on your computer. (Security in depth: sandbox, supply chain and auditing)
- Synthetic voice
- A computer-made voice that reads out loud a text you write. It is useful for voice-overs on short videos, announcements or reading texts aloud. (Creating with AI: images, voice, video and slides)
- Temporary chat
- A conversation that does not use or feed the memory and, in general, does not stay in your history. Not every service has one, and the name changes from one to another. (Projects, custom instructions and memory)
- Terminal
- A program where you talk to the computer in writing: you type a command, press Enter and it answers with text. On Windows it is called Terminal or PowerShell. (The terminal without fear)
- Test account
- An account created only to try out an agent or an automation, with no real data and no access to money. If something goes wrong, the damage stays inside it. (What comes next: agents for non-programmers)
- Throwaway folder
- A folder just for experiments, such as experiencias, where nothing is important. If it goes wrong, you delete it and create another. (Install and sign in: Claude Code and Codex on Windows, Mac and Linux)
- Token
- A piece of a word that the AI works with inside. A short word may be one token; a long or rare one may be split into several. (Why it forgets and why it makes things up)
- Tool (of an agent)
- An action the agent knows how to do, such as reading a file, editing a file or running a command. Without tools, the AI would only write text. (What a coding agent is (and why it is not a chat))
- Training
- The phase in which the model learns patterns from millions of examples. What happens in the world after training, the model does not know by itself. (What AI is (and what it is not))
- Trigger
- The event that makes a workflow start: an email arriving, a set time (every Monday at 9am), a new file in a folder. (Automating your first workflow)
- Two-step verification
- A second proof that it is you signing in, besides the password: a code in a phone app, a message or a security key. Even with the password stolen, the thief cannot get in. (Keeping it alive and safe)
- Typosquatting
- Publishing a malicious package with a name almost identical to a well-known package, waiting for a typing mistake. (Security in depth: sandbox, supply chain and auditing)
- Variable
- A name that holds a value, like a labelled box. Example: preco_papo_seco = 0.25. (Reading code, the minimum: variables, functions and errors)
- Virtual environment (venv)
- A folder, usually .venv, where the Python libraries of one project are kept, separate from those of the rest of the computer. (Asking an agent well: small tasks, done criteria and what a test is)
- Website builder
- An online service where you put the site together from ready-made blocks (text, photos, buttons), without coding. You pay for as long as you use it. (The website: what it needs and how to put it up without coding)
- Workflow
- A YAML file in the .github/workflows folder that tells GitHub Actions when to run and which steps to follow. (CI/CD with agents: GitHub Actions and PR review)
- Workflow (automation)
- A repeated task described in three parts: the trigger that starts it, the steps and the result. For example: an order email arrives, the AI pulls out the details, a row appears in a spreadsheet. (Automating your first workflow)
- WSL
- Windows Subsystem for Linux: a real Linux running inside Windows, installed with the command wsl --install. (Folders, files and an editor (Windows, Mac and Linux))