This glossary covers 36 words you will hear from developers, vendors and in software proposals. Terms run alphabetically in four short groups, and each one gets a plain definition you can repeat to a colleague.
#A-C
Agent (AI agent). A program that uses an AI model to carry out a task, such as answering a message, looking something up or filling in a form. Unlike a fixed script, it decides which steps to take, so it needs clear limits on what it can touch.
API. A defined way for one piece of software to ask another for data or to do something, like the order slip between a diner and a kitchen. When two of your tools “talk to each other,” an API is usually the connection.
Authentication. Proving who you are, usually with a password, a code sent to your phone or a security key. It answers “who is this?” and nothing more.
Authorization. Deciding what a verified person is allowed to do. A clerk and an owner can both log in (authentication), but only the owner may see payroll (authorization).
Automation. A task that runs by itself when something triggers it, such as sending a reminder when an invoice is three days late. Good automations handle the routine cases and pass the odd ones to a person.
Backend. The part of software users never see: the servers, rules and database that store data and do the work. When you press “submit,” the backend receives it.
Backup. A separate copy of your data, kept so you can recover if the original is lost, deleted or corrupted. A backup nobody has tried restoring is only a hope, so ask when the last restore test happened.
Bug. A mistake in the software that makes it behave differently from what was intended. Bugs are normal; what matters is how quickly they are reported, found and fixed.
CI/CD. Short for continuous integration and continuous delivery: an automated line that tests every change and, if the tests pass, prepares it for release. It lowers the chance that a rushed change breaks something.
CRM. Customer relationship management system: software that records your contacts, conversations and deals so the whole team sees the same history. It usually replaces a shared spreadsheet of leads.
#D-H
Database. An organized store of information that software can search and update quickly, such as customers, orders and invoices. Think of a very strict spreadsheet whose rules prevent duplicate or incomplete rows.
Deploy. To put a new version of the software onto the servers where people use it. A well-planned deploy causes little or no interruption.
Encryption. Scrambling data so only someone with the right key can read it. It protects information sitting on a disk (“at rest”) and information moving across the internet (“in transit”).
ERP. Enterprise resource planning system: one platform that handles core operations such as inventory, purchasing, accounting and payroll. Companies often buy one and then connect other software to it.
Frontend. The part of software people see and touch: pages, buttons, forms and screens. It sends requests to the backend and shows what comes back.
Hallucination. When an AI model states something false in a confident, fluent tone, such as an invented policy or a reference that does not exist. It is why answers that matter should be tied to your real data and checked.
#I-P
Integration. A connection that lets two systems share data, for example your booking tool sending new customers to your accounting software. It is usually built on an API or a webhook.
Latency. The delay between asking for something and getting a response. On a phone call with an AI agent, high latency sounds like awkward pauses.
LLM. Large language model: the kind of AI behind chat assistants. It is trained on a very large amount of text, which lets it draft, summarize, classify and answer questions in everyday language.
Migration. Moving data or a whole system from one place to another, such as from spreadsheets into a new database. Developers also use the word for a controlled change to a database’s structure.
MVP. Minimum viable product: the smallest version that does the essential job for real users. It is meant to be used and learned from, not shown off as a rough demo.
Production. The live version of the software that real customers and staff use. Changes here have real consequences, which is why they are tested somewhere else first.
Prompt. The instructions and context you give an AI model, including the question itself. A prompt that states the task, the tone and the limits gives more consistent results than a vague one.
#Q-Z
QA. Quality assurance: checking that software works as promised before and after a release, by testing features, odd cases and past fixes. It works best when someone other than the builder does it.
Repository. The project folder, tracked by version-control software such as Git, that holds the files along with a history of every change. Often shortened to “repo.”
Rollback. Returning software or data to an earlier working version after a release goes wrong. It is the undo button for a bad deploy.
SaaS. Software as a service: software you rent by subscription and use in a browser while the vendor runs it. You pay for access, and the vendor decides what the product does.
Scope. The written list of what a project will and will not include. It sets the price and the expectations, and changes after sign-off should be agreed on paper.
SLA. Service level agreement: a written promise about service, such as how fast a problem gets a response or how much uptime to expect, and what happens if the promise is missed.
Staging. A private copy of the software, set up like production but with test data, where changes are tried out before real users see them.
Tenant (multi-tenant). One customer account inside a system that serves many. In a multi-tenant system everyone shares the same software, but each tenant sees only its own data, like apartments in one building with separate locked doors.
Ticket. A record of one task, request or problem in a tracking tool, with an owner and a status. “Open a ticket” means write it down so it gets tracked instead of forgotten.
Uptime. The share of time a system is available and working, usually stated as a percentage over a month or a year. Ask what counts as “down” and how it is measured.
Webhook. A message one system sends automatically to another when something happens, like a text alert saying “payment received.” It saves the other system from asking over and over whether anything changed.
White-label. Software built by one company and presented under another company’s name, logo and web address, so the end customer never sees the original maker.
Workflow. The sequence of steps that moves a piece of work from start to finish, including who does each step. Writing it out on paper is usually the first step before automating any of it.
