AI and AEO terms,
in plain language.

The words that come up when you put AI agents to work in a small business, defined without the jargon.

Agentic harness
The layer around AI agents that makes them safe to use in a business: it gives them your data and knowledge, limits what each can access, decides what needs a person's approval, and logs every action.
AI agent
An AI system that carries out tasks, not just answers questions. It can read information, decide on steps, use tools such as an ad platform or a store, and report back.
AEO (answer engine optimization)
Making your business the source that AI assistants such as ChatGPT, Perplexity, Gemini, and Google AI cite when people ask them a question. The answer-engine counterpart of SEO.
GEO (generative engine optimization)
Another name for AEO, used especially for visibility in AI-generated answers and summaries.
Answer engine
A tool that replies to a question with a written answer instead of a list of links, often citing its sources. ChatGPT, Perplexity, and Google's AI answers are examples.
SEO (search engine optimization)
Improving a website so it ranks higher in traditional search results. Still the foundation AEO builds on.
Brain
In Humonix, the single store of a business's data and know-how (offers, prices, customers, voice, rules) that every agent works from. It belongs to the business and can be exported.
Approval rule
A rule that decides whether an agent's action ships on its own or waits for a named person. For example: budget changes wait for the owner; weekly reports ship automatically.
Approvals queue
The list of agent proposals waiting for a decision, each with its reasoning and sources. The person can approve, request changes, or reject.
Audit log
A permanent record of what each agent read, what it did, what it cost, and who approved it.
LLM (large language model)
The kind of AI model behind assistants and agents, trained on large amounts of text to read, write, and reason. Claude, GPT, and Mistral models are examples.
Model routing
Choosing a different AI model for each task: a frontier model for hard reasoning, a faster and cheaper one for routine work.
Grounding
Making an AI answer from specific sources, such as your own documents and data, instead of only from what it learned in training.
RAG (retrieval-augmented generation)
A common grounding technique: find the most relevant passages first, then give them to the model to write its answer.
Hallucination
When an AI states something false with confidence. Grounding, cited sources, and approval rules reduce the risk and catch what slips through.
Prompt injection
Text hidden in an email, web page, or document that tries to give an AI agent instructions. A harness treats such content as data, never as orders.
Least privilege
Giving each agent only the access its job needs. The ads agent can change ad budgets within limits; it can't read your accounting.
Schema markup
Structured data added to web pages (from schema.org) that tells search and answer engines exactly what a page is about: a business, a product, a FAQ, an article.
llms.txt
A plain-text file at the root of a website that summarizes the site for AI models and points them to its most useful pages.
Data residency
Where your data is physically stored, for example in Canada or the US. It matters for privacy laws and contracts.

Want to see these ideas running on a real business?