AI glossary for Australian business owners

An AI system that can autonomously plan, execute and iterate on a multi-step task using a language model plus tools. Different from a chatbot, agents act, they don't just respond.

For your business: An agent is what answers an after-hours enquiry, checks your calendar and books the job. On Autopilot's AI Front Desk is an agent; a website widget that only replies to questions is a chatbot.

An AI system that takes actions in the world (reads files, writes responses, books appointments, sends emails) in an autonomous loop, rather than only chatting in response to prompts.

For your business: Agentic means the AI acts inside your systems, so permissions matter. Under the Privacy Act you stay responsible for what it does with customer data: start it read-only and add write access once it has a clean record.

The repeating cycle where an AI agent decides what to do, takes an action, observes the result, and decides what to do next, until the task is complete.

For your business: Every pass through the loop costs tokens, so a badly scoped agent can run up an API bill. Cap the number of steps and log every action so a person can review what it did.

The maximum amount of text (in tokens) a language model can hold in its working memory at once. Modern frontier models support 200k-1M+ tokens of context.

For your business: A bigger window lets an agent read a whole contract, a month of Xero transactions or your full price list in one go. Claude Opus 5.5 and Sonnet 5.5 accept up to 1 million tokens (as at October 2026).

A numerical representation of text (or images, audio) as a vector. Similar meanings produce similar vectors. The foundation of semantic search and most RAG systems.

For your business: Embeddings let an assistant find the right policy or past job note even when a customer uses different words. Most small businesses never touch them directly; they sit inside the search layer of a RAG setup.

Continued training of a pre-trained language model on a smaller dataset of domain-specific examples. Adjusts the model's behaviour without training from scratch.

For your business: Rarely worth it for a small business. Clear instructions, examples of your past replies and retrieval from your own documents give you your voice and your facts for a fraction of the cost.

When a language model emits a structured request to invoke a specific function (with typed arguments), rather than free-form text. The underpinning of tool use.

For your business: This is how an AI fills in a ServiceM8 job, a Xero draft invoice or a calendar booking with the right fields instead of a paragraph of text.

When a language model generates information that's false but stated confidently. The defining failure mode of modern AI. Mitigated, not eliminated.

For your business: The reason a person signs off anything that goes to the ATO, a client or a patient. Ground the AI in your own documents, ask it to cite its source, and check every number before it leaves the business.

When a language model learns from examples provided in the prompt itself, without any weight changes. The trick behind few-shot prompting.

For your business: Pasting five of your best customer replies into the instructions is often all it takes to get an AI writing in your voice. No training run, no extra cost.

Coaxing an AI to produce output it's been trained to refuse. Different from prompt injection; both are security concerns for businesses deploying AI.

For your business: If a public chatbot can be talked into promising a discount or giving advice you would never give, that is a jailbreak. Keep pricing and policy decisions out of the AI's hands and test it before launch.

A neural network trained on enormous amounts of text to predict and generate language. The 'brain' behind Claude, GPT, Gemini and other modern AI products.

For your business: Claude (Anthropic), ChatGPT (OpenAI), Gemini (Google) and Microsoft Copilot are all built on large language models. On Autopilot builds on Claude.

The time between sending a request to an AI model and getting a response. Matters most for user-facing features; not so much for background batch work.

For your business: A phone or chat agent needs to answer in a second or two; an overnight bookkeeping check can take minutes. Smaller, faster models also cost less, so match the model to the job.

An open standard for connecting AI assistants to external tools and data sources. Anthropic's protocol that lets Claude (and others) talk to Xero, Shopify, Slack, etc, in a structured way.

For your business: MCP is how Claude connects to tools like Xero, Shopify and Slack. Xero publishes an official read-only Claude connector and an open-source MCP server (as at October 2026).

A model that accepts more than one type of input (text + images, sometimes audio + video). Modern Claude, GPT, Gemini are all multimodal.

For your business: A multimodal model can read a photo of a receipt, a supplier invoice PDF or the picture of a leaking tap a tenant texts in.

The input you send to a language model. Includes your question or instruction, plus any context, examples, and rules you give the model to work from.

For your business: The prompt is most of the difference between a useful assistant and a generic one: include your business facts, your rules, your tone and an example of a good answer.

A feature that lets language models reuse previously-processed prompt content at a discounted rate, usually ~90% off the input price. Major cost-saver on multi-turn conversations.

For your business: Caching is a big part of why a well-built agent costs a few dollars a month to run. Instructions it rereads on every call are billed at 10% of the normal input price on most current Claude models, and less on Opus 5.5 and Fable 5.1.

A security risk specific to AI systems: untrusted text (in an email, document, or web page) containing hidden instructions that hijack the AI to do something the user didn't intend.

For your business: An email or web form can hide text telling your AI to forward data or ignore its rules. Agents that read customer messages need narrow permissions and must never act on instructions found in the content they read.

A language model trained or configured to think step-by-step before answering, often visibly. Better at hard problems, slower and more expensive than standard models.

For your business: Worth paying for on hard problems such as untangling a messy ledger or pricing a complex quote. Overkill for answering opening-hours questions.

A training technique where humans rank a model's outputs and the model is fine-tuned to prefer the higher-ranked responses. How Claude, GPT and others were taught to be helpful instead of just plausible.

For your business: Nothing you need to do yourself: it is part of how vendors train their models, and it is why models are polite and cautious by default.

A pattern where an AI model retrieves relevant information from a knowledge base before generating its response. Lets AI answer questions about your specific data without retraining the model.

For your business: RAG is how an AI answers from your price list, policies and job history rather than the open internet. It is the standard way to teach AI about your business without retraining a model.

Constraining a language model to produce output that conforms to a specific schema (e.g. valid JSON matching a defined shape). Removes the guesswork from parsing AI responses.

For your business: Structured outputs mean the answer lands as clean fields in your CRM or spreadsheet rather than a paragraph someone has to retype.

The instructions you give a language model upfront, before any user message. Sets the model's role, voice, rules and constraints for the whole conversation.

For your business: In a business agent the system prompt holds your rules: what it may quote, when to hand over to a person, AHPRA or Spam Act limits, and your tone of voice.

The unit of text a language model reads and writes. Roughly 4 characters of English text per token. Models are priced per million tokens.

For your business: Tokens are how AI usage is billed. Claude Sonnet 5.5 costs US$2 per million input tokens and US$10 per million output tokens (as at October 2026), and a million tokens is very roughly 600,000 to 750,000 English words.

When a language model calls an external function or API as part of generating its response. The foundation of agentic AI, turns the model from a text generator into a system that can act.

For your business: Tool use is what lets AI check stock in Shopify, look up a client in Cliniko or create a draft invoice, instead of only telling you how to do it.

A database optimised for storing and searching numerical representations (embeddings) of text, images, or other data. Used to give AI systems memory and search over large document collections.

For your business: Useful once you have thousands of documents (job notes, contracts, support tickets) to search. Most small businesses do not need a separate one.