When you start looking into AI automation, you quickly run into vocabulary that seems reserved for developers: API, webhook, trigger, prompt, workflow. Yet behind these words lie simple ideas, ones you already use without knowing it.
This glossary does not aim to be exhaustive or academic. It is organised by situation: what you come across when you describe a task, when you connect your tools, when the agent runs, and when you monitor its results. For each term, a short definition and a concrete example. If you are looking for brief definitions in alphabetical order, see the site glossary; this article explains how these ideas fit together in practice.
When you describe the task#
AI agent#
A program that carries out a task from start to finish, without you having to step in at each stage: it reads, understands, decides according to your rules, acts in your tools and reports back to you.
Example: an agent that reads your emails every morning, spots quote requests and prepares a reply for each.
Conversational assistant#
A tool you chat with (like ChatGPT). It answers when you talk to it, but does nothing on its own between two conversations. That is the main difference from an agent. Our article ChatGPT, Claude, Mistral: which AI model should power an agent? explains this distinction in detail.
Brief (or instructions)#
The description of what you expect from the agent: what, when, with which tools, in what form, with which exceptions. It is the element that most influences the quality of the result.
Example: "Every Friday at 4 pm, send me the list of quotes sent more than ten days ago with no reply, with the amount and the contact."
Prompt#
The instruction given to an AI model. In everyday language, "prompt" and "brief" are often used interchangeably. On an agent platform, you write a brief; the platform turns it into detailed instructions for the model.
Architect#
On Future Agent, the architect is the AI that designs your agent from your brief: it asks the missing questions, identifies the tools to connect, shows you a sample result and calculates the monthly price. You talk to the architect; the agent does the work afterwards.
Model and template: two words not to confuse#
Two neighbouring words come up often in this field:
- language model: the AI engine (GPT, Claude, Gemini, Mistral…) that understands and writes;
- agent template: a ready-to-use task, with its brief and steps, that you adapt to your situation. The library offers more than 300, which you can filter by sector and profession.
Rule (or guardrail)#
An instruction the agent must always follow, often phrased with "always" or "never".
Example: "Never chase an invoice marked as disputed."
When you connect your tools#
Connector#
The link between the agent and one of your tools (Gmail, Outlook, Google Drive, QuickBooks, HubSpot, Slack…). Without a connector, the agent can neither read your data nor act for you. Future Agent gives access to more than 1,600 apps.
One-click connection (OAuth)#
The method that lets you authorise the agent to use a tool without giving it your password. You click "Connect", the tool asks you to confirm, and you are done. You can withdraw this authorisation at any time. The technical term is OAuth; you have already used it if you have ever signed in to a website "with Google".
API#
An entry point that a piece of software opens to other software so they can exchange data. When people say a tool "has an API", it means an agent can talk to it directly. You do not need to know how to use it: connectors take care of that.
Permissions (or access scope)#
What the agent is allowed to do in a tool: read only, or also write, send, delete. The right principle: only grant the permissions the task needs.
When the agent runs#
Trigger#
What starts the agent. There are three main families:
| Type of trigger | Principle | Example |
|---|---|---|
| Scheduled | At a set time or regular interval | Every Monday at 8 am |
| Event | When something happens | When an email with an invoice arrives |
| On demand | When you launch it yourself | A "Run" button in the dashboard |
On Future Agent, an agent can also be triggered by another agent, when that agent finishes its work.
Scheduling (or cron)#
The way you define when a scheduled agent runs. "Cron" is the technical name for a format used in computing to write schedules. You do not have to write it: you say "Monday to Friday at 9 am" and the platform handles the rest.
Webhook#
A message a tool sends automatically when an event occurs, to notify another tool. It is one of the mechanisms behind event triggers.
Example: your online store sends a webhook for every new order, and the shipment tracking agent starts.
Run#
One occasion when the agent does its work, from start to finish. An agent scheduled every morning does one run a day.
Capabilities#
The types of actions an agent can perform: producing files (PDF, Excel, Word), searching the web, sending emails, browsing a website, running a calculation. The architect chooses the capabilities your task needs.
Deliverable#
What the agent produces at the end of a run: an email, a PDF report, an Excel spreadsheet, a Word document, an update in a CRM, a website.
When agents work together#
Workflow (or agent chain)#
Several linked agents that hand work over to one another, each specialised in one step.
Example: one agent finds prospects, a second prepares a first message, a third follows up with those who have not replied. The prospecting, contact and follow-up chain works this way.
Shared table#
A data table that one agent fills in and another reads. It is the meeting point between two agents in a chain: the downstream agent only receives the rows added since its last pass.
Upstream and downstream agents#
In a chain, the upstream agent works first and passes things on; the downstream agent receives them and carries on.
Memory and context#
Context#
All the information available to the agent when it works: your brief, the run's data, its memory, your reference documents. The more relevant the context, the better the result.
Individual memory#
What an agent remembers from one run to the next: your preferences, useful facts, summaries of its past runs.
Example: the agent remembers that a particular customer prefers to be contacted in the morning.
Collective memory#
Information shared between all the agents of the same account (never with another account): company details, common rules, house tone.
Context files#
Documents you give an agent to refer to: price list, terms and conditions, sample messages, brand guidelines. They are split up and indexed so the agent finds the relevant passage at the right moment.
Document search (RAG)#
The mechanism that lets an agent find the passages in your documents that are relevant to the task at hand, instead of rereading everything. RAG stands for "retrieval-augmented generation".
When you monitor the results#
Run log#
The detailed account of what the agent did during a run: what it read, the actions taken, the deliverable produced, any errors. It is your first port of call when in doubt.
Human approval (or human in the loop)#
A step where the agent stops and waits for your approval before acting. Essential for irreversible actions: payment, sending a contract, a purchase. Often called human in the loop.
Hallucination#
When an AI model produces false information presented with confidence: an invented figure, a reference that does not exist. The risk decreases when the agent works from your real data and clear rules, but it never disappears completely. Hence the value of reviews at the start and approvals on sensitive topics.
Extraction#
Reading a document (invoice, receipt, delivery note, email) and pulling structured information out of it: amount, date, supplier, order number.
Text recognition (OCR)#
The technology that reads the text in an image or a scanned PDF. It lets an agent read the photo of a receipt, for example.
When the agent builds a website#
Project agent#
An agent that builds a lasting deliverable, such as a website, rather than a recurring task. On Future Agent, you see the preview in the chat, ask for changes, and then the site is hosted.
Preview#
A viewable version of the site before it goes live, so you can check it and ask for adjustments.
Hosting#
The service that keeps your site online at all times. For a project agent, it includes a number of revisions per month, with higher tiers if you need more. Details are on the pricing page.
Using this glossary day to day#
You do not need to memorise all these terms to use agents. Three are enough to get started: brief, trigger and connector. The others will come naturally as you add agents and link them together.
To put it into practice, the 30-day plan to get started with AI uses these ideas step by step, and the features page shows how they translate into the tool.