The Jargon, Translated.

Agentic

The AI works out the steps itself instead of following a fixed script. Genuinely powerful, which is exactly why it needs rules and supervision.

AI agent

Software that can hold a conversation and take actions — look up availability, book a slot, update a record — rather than only producing text. The actions are what separate it from a chatbot.

API

The socket one piece of software exposes so another can plug into it. When someone says two tools "integrate", they usually mean one is using the other's API.

Augmentation

You still do the work — AI just makes you far quicker at it. Drafting, summarising, first passes. Most teams should start here.

Automation

The software does the whole task: a lead comes in, the reply goes out, the CRM updates. Nobody touched anything.

Chatbot

Older, narrower ancestor of an AI agent: a decision tree of pre-written replies. If it cannot do anything except talk, and it breaks the moment someone phrases a question unusually, it is a chatbot.

Context window

How much text a model can hold in mind at once — the conversation so far, plus whatever you have given it to work from. Exceed it and the earliest details quietly drop off.

CRM

The system of record for your customers and deals. For most businesses it is the thing that is meant to be up to date and never is, which is why writing to it automatically matters more than it sounds.

Guardrails

The explicit rules about what a system may and may not do or say, and when it must hand over to a person. The unglamorous part that decides whether any of this is safe to point at customers.

Hallucination

When a model states something false with complete confidence. It is a property of how these systems work, not a bug that gets fixed — which is why grounding and guardrails exist.

Human in the loop

A person reviews or approves before something goes out. Standard practice while a new system beds in, and permanent for anything high-stakes.

Integration

Getting two systems to exchange information without a person retyping it. Most of the value in an automation project is here rather than in the AI itself.

LLM (large language model)

The engine underneath tools like ChatGPT and Claude. It predicts text, which is why it is superb at language and needs checking on facts.

Missed-call text-back

A named, well-established tactic: when a call goes unanswered, the caller automatically gets a text within seconds so the conversation continues instead of ending at voicemail.

Orchestration

Getting the tools you already pay for talking to each other, so information moves without being retyped. Less exciting than agents; usually worth more.

Prompt

The instructions you give a model. A good one specifies the role, the context, the format you want back and what to do when unsure — which is most of the skill.

RAG (retrieval-augmented generation)

Giving the model your actual documents to answer from, rather than relying on what it absorbed in training. This is how an agent quotes your real pricing instead of inventing plausible pricing.

Speed to lead

How long it takes you to respond to a new enquiry. Widely studied outside this agency, and consistently the cheapest thing most businesses can fix.

Token

The unit models read and bill in — roughly three-quarters of a word. Relevant mainly because it is what usage costs are measured in.

Webhook

A message one system fires at another the instant something happens, instead of the second system having to keep asking. It is what makes "within seconds" possible.

Workflow automation

Automating an internal sequence — quote to invoice, booking to dispatch, form to report — rather than a customer conversation. Less visible than an agent, often a bigger saving.

Knowing the words is the shallow end. Knowing which of these belongs in your business — and wiring it together so it keeps running without you — is the work we do every day.