AI model
- In plain English
- A trained computer system that produces or classifies information from patterns in data.
- Why it matters
- It affects what a tool can do, how it behaves and what needs checking.
- Example
- Choosing a model is like choosing a vehicle: a ute and a bus can both travel, but suit different loads.
Related terms: large language model, generative AI
Read next: Guide 009, Guide 048
large language model
- In plain English
- A type of AI model trained to work with language, including questions, summaries and drafts.
- Why it matters
- It helps you understand why a chat tool can write fluent text without independently proving it.
- Example
- A large language model can draft a note from supplied points, then you check the points against the source.
Related terms: AI model, prompt, hallucination
Read next: Guide 009, Guide 012
generative AI
- In plain English
- AI that creates new output such as text, images, code or audio from instructions and context.
- Why it matters
- It is useful for drafts and options, but the new output still needs human review.
- Example
- A generated poster draft can suggest a layout; it does not confirm the event date is correct.
Related terms: AI model, prompt, human in the loop
Read next: Guide 001, Guide 051
prompt
- In plain English
- The instructions and material you give an AI tool for one task.
- Why it matters
- A clearer prompt usually produces a more reviewable first draft.
- Example
- “Use these meeting notes to list confirmed actions and questions” is a prompt.
Related terms: context window, system instruction, token
Read next: Guide 002, Guide 013
system instruction
- In plain English
- A higher-level instruction that guides an AI system’s behaviour for a conversation or tool.
- Why it matters
- It can shape what the tool is asked to do, but it does not remove your responsibility to check results.
- Example
- A system instruction might say to mark missing facts rather than invent them.
Related terms: prompt, guardrail, agent
Read next: Guide 014, Guide 038
token
- In plain English
- A small unit of text that an AI system processes; a word can be one or several tokens.
- Why it matters
- Token limits affect how much material and output a tool can handle at once.
- Example
- A long report may need to be split into sections because it uses more tokens than a tool can consider.
Related terms: context window, large language model
Read next: Guide 009, Guide 021
context window
- In plain English
- The amount of text, images or other information an AI model can consider in one request.
- Why it matters
- It helps you plan how to supply source material without assuming the tool has seen everything.
- Example
- If a report is longer than the context window, work through it in labelled sections and retain references.
Related terms: token, prompt, memory
Read next: Guide 013, Guide 021
hallucination
- In plain English
- A confident-looking AI output that is inaccurate, unsupported or invented.
- Why it matters
- Knowing this term reminds you to verify important claims before relying on them.
- Example
- A draft that invents a delivery date from incomplete notes is a hallucination.
Related terms: grounding, human in the loop, large language model
Read next: Guide 005, Guide 022
multimodal
- In plain English
- Able to work with more than one type of information, such as text, images, sound or documents.
- Why it matters
- It changes what input you may need to review for privacy, accuracy and permissions.
- Example
- A multimodal tool may read a photo of a whiteboard as well as typed meeting notes.
Related terms: AI model, context window
Read next: Guide 004, Guide 051
memory
- In plain English
- Information an AI product may retain or reuse between interactions, depending on its settings.
- Why it matters
- It affects what you should check in privacy settings and what context may carry into later work.
- Example
- If memory is enabled, a tool may remember a writing preference until you remove or change it.
Related terms: context window, project, system instruction
Read next: Guide 010, Guide 011
project
- In plain English
- A named workspace that groups related files, instructions or conversations in a tool.
- Why it matters
- It can help keep a task organised, provided its contents and access are appropriate.
- Example
- A project for a campaign can hold approved copy and a review checklist in one place.
Related terms: memory, context window, workflow
Read next: Guide 006, Guide 052
connector
- In plain English
- A connection that lets an AI application work with another service or information source.
- Why it matters
- It may expose account data or actions, so its permissions deserve review.
- Example
- A calendar connector can make meeting data available to an AI app if you approve the access.
Related terms: plugin, MCP, API
Read next: Guide 034, Guide 040
plugin
- In plain English
- An installed add-on that gives an application extra capability, tools or integrations.
- Why it matters
- A plugin can request data access or take actions, so check its source and permissions first.
- Example
- A plugin that creates tickets should be tested with safe data before it sees real customer requests.
Related terms: connector, skill, API key
Read next: Guide 034, Guide 037
skill
- In plain English
- Reusable instructions that help an AI carry out a particular kind of task consistently.
- Why it matters
- A good skill makes a repeatable task clearer while keeping limits and checks visible.
- Example
- A weekly-update skill can require source notes, headings and an operator check.
Related terms: workflow, agent, plugin
Read next: Guide 034, Guide 035
workflow
- In plain English
- A defined sequence of steps, inputs, checks and outputs for a repeated task.
- Why it matters
- It helps you decide whether work is stable enough to repeat or automate.
- Example
- A workflow can turn approved notes into a draft update and send it to a reviewer.
Related terms: automation, skill, agent
Read next: Guide 006, Guide 039
automation
- In plain English
- A system that performs a defined step or sequence after a trigger, often without a person doing each step manually.
- Why it matters
- It can save time only after the underlying process is reliable and its exceptions are understood.
- Example
- A workflow may automatically create a reminder when an approved form is submitted.
Related terms: workflow, agent, human in the loop
Read next: Guide 047, Guide 032
agent
- In plain English
- An AI system that can plan or take several steps toward a goal using available tools.
- Why it matters
- Agents need clear limits because their actions can affect systems beyond a single draft.
- Example
- An agent might gather approved files and prepare a report, then wait for a human before sending anything.
Related terms: subagent, workflow, guardrail
Read next: Guide 038, Guide 039
subagent
- In plain English
- A smaller agent assigned a bounded part of a larger task by another agent or system.
- Why it matters
- It helps split work, but accountability and permissions still need to be clear.
- Example
- A research agent may ask a subagent to list source dates while it prepares the decision brief.
Related terms: agent, workflow, human in the loop
Read next: Guide 038, Guide 041
MCP
- In plain English
- A shared way for an AI application to discover and use tools or information supplied by another system.
- Why it matters
- It can let one compatible connection work across several AI applications, while the connected service still controls access.
- Example
- Think of it as a standard plug shape. The plug makes connection possible; it does not decide which rooms you may enter.
Related terms: connector, MCP server, API
Read next: Guide 034, Guide 036
MCP server
- In plain English
- A service that exposes tools, prompts or information through the Model Context Protocol.
- Why it matters
- It tells you where capabilities come from and what service needs a permissions review.
- Example
- A calendar MCP server can offer a “list events” tool to a compatible AI application.
Related terms: MCP, connector, API
Read next: Guide 036, Guide 037
API
- In plain English
- A defined way for software systems to request information or actions from each other.
- Why it matters
- APIs often sit behind integrations, so their access and error handling matter.
- Example
- A booking system API may return available times to an internal tool.
Related terms: API key, connector, MCP
Read next: Guide 034, Guide 040
API key
- In plain English
- A secret credential that lets software prove it is allowed to use an API.
- Why it matters
- Anyone holding an active key may be able to use its permitted access, so it must be stored and revoked carefully.
- Example
- Treat an API key like a spare keycard with a record of which doors it can open.
Related terms: API, connector, guardrail
Read next: Guide 037, Guide 040
RAG
- In plain English
- Retrieval-augmented generation is a pattern where an AI system looks up relevant source material before producing an answer.
- Why it matters
- It can improve traceability when the retrieved sources are visible and still checked.
- Example
- A policy assistant retrieves the current policy section before drafting an answer for review.
Related terms: grounding, embedding, vector database
Read next: Guide 022, Guide 052
embedding
- In plain English
- A numerical representation of meaning that helps software find similar pieces of information.
- Why it matters
- It is often used behind search and RAG, where the result still depends on the source material and settings.
- Example
- Two notes about annual leave may be close in embedding space even when they use different wording.
Related terms: RAG, vector database, inference
Read next: Guide 022, Guide 035
vector database
- In plain English
- A database designed to store and search embeddings so systems can find semantically related material.
- Why it matters
- It affects how a RAG system finds source passages and what needs access controls.
- Example
- A vector database can help a policy tool find the paragraph most related to a leave question.
Related terms: embedding, RAG, API
Read next: Guide 022, Guide 052
fine-tuning
- In plain English
- Further training of a model on selected examples to change how it performs a task.
- Why it matters
- It is a substantial data and governance choice, not the first fix for a vague prompt.
- Example
- A team may fine-tune a model only after testing prompts and examples for a stable, high-volume task.
Related terms: training data, inference, prompt
Read next: Guide 035, Guide 031
training data
- In plain English
- Material used to train or adjust an AI model before it is used for a task.
- Why it matters
- Its origin, quality and permissions affect bias, privacy and suitability.
- Example
- If historical examples contain inconsistent labels, a model can learn those inconsistencies.
Related terms: fine-tuning, AI model, grounding
Read next: Guide 029, Guide 031
inference
- In plain English
- The process of using a trained AI model to produce an output from a prompt or input.
- Why it matters
- It separates the act of using a model from the earlier work of training it.
- Example
- When you ask a model to summarise notes, that response is produced during inference.
Related terms: AI model, temperature, token
Read next: Guide 009, Guide 013
temperature
- In plain English
- A setting that can influence how varied or predictable some AI model outputs are.
- Why it matters
- It may affect draft variation, but it does not make an answer more factual.
- Example
- A lower setting may produce more repeatable wording for a standard summary.
Related terms: inference, hallucination, prompt
Read next: Guide 013, Guide 017
grounding
- In plain English
- Giving an AI system relevant source material or constraints so its output can be checked against them.
- Why it matters
- Grounding helps keep work tied to evidence instead of plausible-sounding guesses.
- Example
- A grounded draft uses the supplied policy text and points back to its sections.
Related terms: RAG, hallucination, human in the loop
Read next: Guide 005, Guide 022
guardrail
- In plain English
- A limit or check designed to prevent an AI system from taking unsafe, unauthorised or unsuitable actions.
- Why it matters
- Guardrails help define what happens when a task crosses a boundary or becomes uncertain.
- Example
- A guardrail can require human approval before an agent sends an external email.
Related terms: agent, human in the loop, system instruction
Read next: Guide 038, Guide 041
human in the loop
- In plain English
- A person who reviews, approves or corrects an AI-assisted process at a meaningful point.
- Why it matters
- It keeps accountability with someone who can assess context, consequences and exceptions.
- Example
- A person checks evidence and approves a customer reply before it is sent.
Related terms: guardrail, agent, hallucination
Read next: Guide 005, Guide 041
Cowork
- In plain English
- Claude’s task experience for working with the files and tools you permit.
- Why it matters
- Your interface may combine chat and Cowork; check the current guide.
- Example
- Select a folder, then ask for a sorting plan.
Related terms: agent
Read next: Guide 032
Artifact
- In plain English
- A reusable output such as a document or small interactive tool.
- Why it matters
- You still need to check the output against its inputs.
- Example
- Two versions of a weekly brief.
Related terms: project
Read next: Guide 025
Global instructions
- In plain English
- Standing preferences for how the assistant should respond.
- Why it matters
- Keep task facts out of preferences that apply broadly.
- Example
- Use Australian English and concise answers.
Related terms: system instruction
Read next: Guide 018
Browser access
- In plain English
- Permission for an assistant to navigate websites.
- Why it matters
- This differs from connecting an email or file service.
- Example
- Compare an item on three seller pages.
Related terms: connector
Read next: Guide 040
GitHub
- In plain English
- A service for storing project files and their change history.
- Why it matters
- Files uploaded there leave your device; choose visibility deliberately.
- Example
- Create a private practice repository.
Related terms: project
Read next: Guide 045
Repository
- In plain English
- A project’s files and tracked history.
- Why it matters
- Keep credentials and private data out of practice repositories.
- Example
- A folder with README and sample notes.
Related terms: GitHub
Read next: Guide 045
Commit
- In plain English
- A recorded set of file changes with an explanatory message.
- Why it matters
- History lets you see what changed between versions.
- Example
- Add first weekly brief.
Related terms: Repository
Read next: Guide 042
Private repository
- In plain English
- A repository restricted to people granted access.
- Why it matters
- Private still means hosted online, not stored only on your laptop.
- Example
- A practice repo only you can open.
Related terms: Repository
Read next: Guide 045
README
- In plain English
- A project’s introduction and usage instructions.
- Why it matters
- It helps someone understand and use the files.
- Example
- Purpose, file list and a first exercise.
Related terms: Repository
Read next: Guide 042
Markdown
- In plain English
- Plain text with simple formatting marks.
- Why it matters
- Many instructions and project files use this format.
- Example
- # creates a heading.
Related terms: README
Read next: Guide 035
SKILL.md
- In plain English
- The instruction file describing a reusable skill and when to use it.
- Why it matters
- Inspect the contents before installing a downloaded skill.
- Example
- The weekly-brief sample has instructions and no scripts.
Related terms: skill
Read next: Guide 035
Refining
- In plain English
- Giving specific feedback and checking the revised result.
- Why it matters
- A precise correction is more useful than make it better.
- Example
- Preserve proposed dates rather than confirming them.
Related terms: prompt
Read next: Guide 017
Sync
- In plain English
- Keeping copies up to date across locations.
- Why it matters
- A browser upload is not ongoing automatic synchronisation.
- Example
- Upload a new version after editing locally.
Related terms: Repository
Read next: Guide 045