Free resource
The words yourvendor won’t define.
76 terms and acronyms from software and AI, written for the person signing the invoice rather than the person writing the code. What each one means, why it exists, and where you’ll actually hear it.
You don’t have a technology problem. You have a vocabulary problem, and vocabulary is finite — sixty-odd words carry almost every conversation in this field. Not one of them is conceptually hard.
- API Application Programming Interfaceweb
- A defined list of things one system will do for another, and the exact shape of the request required to ask.
- endpoint web
- One specific address on an API that does one specific thing.
- JWT JSON Web Token, said "jot"identity
- A small packet of facts about a user — their ID, their role, their company — cryptographically signed so it cannot be edited without detection.
- claim identity
- One fact stored inside a token. "Role: manager" is a claim.
- sign / verify identity
- Signing stamps data with a secret key. Verifying proves the stamp is genuine and nothing was altered.
- public / private key identity
- A matched pair. The private one signs; the public one verifies. Publishing the public key is safe and expected.
- UUID Universally Unique Identifierdatabase
- A 36-character random ID, like 8f14e45f-ceea-467a-9e2c-1d2a4b6c8e10. Two systems can generate them independently and never collide.
- session identity
- The system's memory that this particular person is currently signed in.
- cookie identity
- A small piece of data a site asks your browser to hold and send back on every future request.
- SSO Single Sign-Onidentity
- One login that opens every tool your company uses, instead of a separate password per app.
- OAuth identity
- The standard handshake behind "log in with Google." The other service never sees your password — it receives a limited token instead.
- 401 vs 403 web
- 401 means "I don't know who you are." 403 means "I know exactly who you are, and the answer is no."
- TTL Time To Liveinfra
- How long something stays valid before it's thrown away — a login token, a cached page, a DNS record.
- MFA / 2FA Multi-Factor / Two-Factor Authenticationidentity
- Requiring a second proof beyond a password — a code, an app prompt, a hardware key.
- SQL Structured Query Language, said "sequel"database
- The language for asking a database questions. SELECT reads, INSERT adds, UPDATE changes, DELETE removes.
- Postgres PostgreSQLdatabase
- A database — the program that stores your records and answers questions about them.
- schema database
- The shape of your data — which tables exist, what fields they have, and what kind of value each one accepts.
- row / column database
- A row is one record — one customer, one job. A column is one field on every record — every customer's phone number.
- query database
- One question asked of the database.
- transaction database
- A group of database changes that all succeed together or all get undone together. Nothing lands halfway.
- migration database
- A numbered file, saved in version control, that changes the database's structure. Applied in order, never edited after it ships.
- RLS Row Level Securitydatabase
- The database attaching its own filter to every read and write, no matter what the application asked for.
- policy database
- One row-level security rule. "When reading jobs, only return the ones belonging to the caller's company."
- index database
- A lookup structure that lets the database find matching records without reading every single one.
- SQL injection database
- An attack where text typed into a form gets treated as database commands instead of as data.
- tenant database
- One customer in a system that serves many, whose data must never touch another customer's.
- backup vs replica database
- A backup is a copy from a point in time you can restore from. A replica is a live second copy kept in sync right now.
- ORM Object Relational Mapperdatabase
- A library that writes database queries for you from ordinary code.
- request / response web
- The whole of the web. Something asks, something answers, and then both sides forget it happened.
- HTTP verbs web
- GET reads. POST creates. PUT and PATCH update. DELETE removes.
- CRUD Create, Read, Update, Deleteweb
- The four things every business application does to data. Most software is CRUD with opinions on top.
- JSON JavaScript Object Notationweb
- The standard text format for passing structured data between systems. Curly braces and label-value pairs.
- status code web
- The three-digit verdict on every web response. 200 is fine, 3xx moved, 4xx you made a mistake, 5xx we made a mistake.
- client-side web
- Code running on the user's own phone or laptop. Visible to them, editable by them, and therefore untrusted.
- server-side web
- Code running on the company's own machines. Never downloaded, never visible to the user.
- bundle web
- The package of code compiled and shipped down to a visitor's browser.
- SSR / CSR Server-Side / Client-Side Renderingweb
- With SSR the server sends a finished page. With CSR it sends a blank page plus instructions for the browser to draw it.
- SPA Single Page Applicationweb
- A site that loads once and then swaps content in place instead of reloading whole pages.
- PWA Progressive Web Appweb
- A website that installs to a phone's home screen and keeps working with no signal.
- polling web
- Asking a server "anything new yet?" on a repeating timer.
- webhook web
- The reverse of polling. Instead of you asking repeatedly, the other system calls you the moment something happens.
- CORS Cross-Origin Resource Sharingweb
- The browser rule that a page on one website can't freely call another website's API unless invited.
- cache web
- A saved copy of something expensive, handed out instead of recalculated.
- CDN Content Delivery Networkinfra
- Copies of a site's files parked in data centers worldwide so visitors load from one nearby.
- rate limit web
- A cap on how many requests one caller can make in a window. Exceed it and you get refused until it resets.
- DNS Domain Name Systeminfra
- The phone book that turns a domain name into the numeric address of a server.
- TLS / HTTPS Transport Layer Securityinfra
- The encryption wrapping traffic between a browser and a server. The padlock in the address bar.
- load balancer infra
- The traffic director out front, spreading incoming visitors across however many copies of an application are running.
- container infra
- An application plus its exact dependencies, sealed in a box that runs identically on any machine.
- environment variable infra
- A configuration value handed to an application when it starts, instead of being written into the code.
- secret infra
- Any value that grants access — an API key, a database password, a signing key.
- CI/CD Continuous Integration / Continuous Deploymentinfra
- Automation that tests every change and ships the ones that pass.
- staging vs production infra
- Staging is the rehearsal copy. Production is the one real customers touch.
- deploy infra
- The act of putting new code onto the live system.
- rollback infra
- Putting the previous version back after a bad deploy.
- cron infra
- A scheduler that runs a task on a fixed timetable — nightly, hourly, every Monday.
- uptime / SLA Service Level Agreementinfra
- The percentage of time a system is available, and the contractual promise attached to it.
- logs infra
- The running written record of what a system did and when.
- LLM Large Language Modelai
- The model itself — Claude, GPT, Gemini. A program that predicts the next chunk of text given everything before it.
- token ai
- A chunk of text, roughly three quarters of a word. Models read, write, and are billed in tokens.
- context window ai
- The total amount of text a model can hold in mind at once — your input plus its answer.
- prompt ai
- Everything sent to the model: the instructions, the data, and the question.
- system prompt ai
- Standing instructions set before a conversation starts, defining the assistant's role, rules, and tone.
- inference ai
- One run of a model producing an answer. It's what you pay for, per call.
- RAG Retrieval Augmented Generationai
- Search your own documents first, paste only the relevant passages into the prompt, then ask the question.
- embedding ai
- A block of text converted into a long list of numbers that encodes its meaning. Similar meanings produce similar numbers.
- vector database ai
- A database built to find the closest embeddings quickly.
- chunking ai
- Splitting long documents into passages before storing them for retrieval.
- agent ai
- A model given tools and a goal, working in a loop — decide, act, check the result, decide again — until the job is done.
- tool call also called function callingai
- The model asking your software to run something — look up an order, send an email, query a report — and hand back the result.
- MCP Model Context Protocolai
- A standard shape for exposing your tools and data to any AI assistant, so any model can use them without custom wiring.
- fine-tuning ai
- Further training a model on your own examples so it absorbs a specific style or task.
- hallucination ai
- A model producing something fluent, confident, and completely false.
- temperature ai
- A dial controlling randomness. Low is repeatable and literal. High is varied and creative.
- eval evaluationai
- A test suite for AI output — a fixed set of inputs with graded expectations, run after every change.
- human in the loop ai
- A required person-approves-it step before an AI action takes effect.
Why it existsSo two systems can work together without either knowing how the other is built. Everything behind an API can change as long as the shape stays the same.
"Does your software have an API?" means: can other tools talk to it automatically, or is someone retyping data by hand?
Why it existsAn API is the whole menu. An endpoint is a single item on it.
A developer saying "I'll build you an endpoint for that" means one new, narrow capability — usually hours, not weeks.
Why it existsIt lets a system trust a claim about you without looking you up in a database on every single click. Almost entirely a speed decision.
Any modern login uses one. If a vendor can't tell you how long theirs stays valid, that's worth pushing on.
Why it existsBundling facts into the token means the system can make permission decisions instantly instead of querying a database first.
Why it existsIt is a wax seal. Anyone can read a sealed letter, but only the holder of the seal can create one.
Why it existsLets anyone in the world confirm a message came from you, while only you can produce one.
Why it existsEmails change and names repeat. An ID never moves, so it's what the database actually keys on.
Why it existsThe web forgets you between every single request. A session is the thread that survives from one click to the next.
Why it existsIt's a coat check ticket. Your browser carries it automatically, which is why you don't log in again on every page.
Why it existsFewer passwords means fewer reused passwords, and one switch turns off a departing employee's access everywhere at once.
Worth asking any vendor. "Do you support SSO?" tells you a lot about how seriously they take security.
Why it existsLets you prove who you are to a company without handing them a credential they'd then be responsible for protecting.
Why it existsTwo very different problems. Confusing them sends a developer debugging the wrong half of the system.
If your team reports "it says 403," the login worked and a permission is missing. That's a settings problem, not a bug.
Why it existsNothing should be trusted forever. A short TTL means a revoked permission takes effect sooner.
Why it existsPasswords leak in bulk. A second factor means a stolen password on its own is worthless.
The single highest-value security change most small companies can make in an afternoon.
Why it existsOne language that nearly every database in the world speaks, essentially unchanged since the 1970s.
Why it existsFree, extremely reliable, and the default serious choice. If a vendor says "we run on Postgres," that's a good sign, not a red flag.
Why it existsThe database enforces the shape, so malformed data can't get in through a side door.
Why it existsSpreadsheet vocabulary, and the analogy holds almost perfectly.
Why it existsEverything an app displays started as a query. "The report is slow" almost always means "one query is slow."
Why it existsMoney moving between two accounts must not stop in the middle. Same principle protects an order, an invoice, or a job record.
Why it existsSo the test copy and the live copy have identical structure, and anyone can answer "when did this field appear, and who added it?"
If a vendor changes your database by clicking around in an admin panel instead, ask how they'd undo it.
Why it existsBecause applications forget. One missing filter in one place is how one customer sees another customer's data. Enforcing it in the database means the app cannot make that mistake.
The right question for any vendor holding data for multiple customers: "Where is tenant isolation enforced — in the app or in the database?"
Why it existsThe rule lives next to the data instead of being scattered across a hundred different screens and reports.
Why it existsThe difference between a search taking four milliseconds and four seconds. Most "the system got slow as we grew" complaints are a missing index.
Why it existsThe classic version is a name field containing a command that deletes a table. It's decades old, completely preventable, and still in the top ten breaches every year.
Why it existsOne application, one database, many companies. It's how nearly all business software is built — and the isolation between tenants is the whole ballgame.
"Multi-tenant" means you share infrastructure with other customers. That's normal and fine. How they separate you is the question.
Why it existsThey solve different disasters. A replica survives hardware failure. Only a backup survives someone deleting the wrong thing on Tuesday.
"We have replication" is not the same as "we have backups." Ask for both, and ask when a restore was last tested.
Why it existsConvenience and consistency. The tradeoff is that it can hide an inefficient query until your data grows.
Why it existsEverything else — logins, sessions, cookies — exists to add memory to a system that has none by default.
Why it existsA shared convention, so any developer can guess what a piece of a system does before reading a line of it.
Why it existsUseful for scoping. "It's CRUD plus a PDF export and a nightly email" tells a builder almost everything they need to quote it.
Why it existsReadable by a human, parseable by every programming language. It won because it was the simplest thing that worked.
Why it existsThe 4xx versus 5xx split is the fastest triage in software: your fault or theirs.
"It's throwing 500s" means the vendor's system is broken, not your team's usage.
Why it existsAnything checked here can be skipped by a determined user. It's for convenience, never for enforcement.
If a vendor says a permission is "handled in the interface," that's not a security control. Ask where it's enforced on the server.
Why it existsThe only place secrets and real decisions belong, because it's the only place a user cannot reach.
Why it existsAnything in the bundle is public, whether you meant it to be or not. It's the single most common way an API key leaks.
Why it existsSSR shows up faster and is visible to Google. CSR feels more app-like once loaded. Most modern sites mix both.
If a site's content is invisible in search results, rendering strategy is the first thing to check.
Why it existsFeels like software rather than a website. Costs a heavier first load in exchange.
Why it existsApp-store reach without app-store approval, review delays, or a 30% cut. Often the right answer for an internal field tool.
Why it existsThe simple alternative to a live connection. Costs bandwidth, buys reliability — and reliability usually wins for business software.
Why it existsInstant instead of delayed, and far cheaper than checking every thirty seconds forever.
The backbone of most automation. "Does it support webhooks?" decides whether two tools can be wired together cleanly.
Why it existsIt stops a malicious page from quietly reading data you're logged into somewhere else.
Why it existsSpeed. It's also the number one reason someone says "I published the change and the site looks exactly the same."
There are three separate caches in play — the browser's, the delivery network's, and the app's. Naming which one is most of the fix.
Why it existsPhysics. A round trip from Indiana to Virginia has a floor that no amount of clever code can beat.
Why it existsProtects a system from both abuse and an accidental infinite loop in somebody's script.
Ask about limits before building an integration. Discovering them in production is an expensive way to learn.
Why it existsHumans remember names; networks route numbers.
"It's a DNS issue" usually means the site is fine and the directions to it are wrong. Changes can take hours to spread.
Why it existsWithout it, anyone on the same coffee-shop wifi can read the session cookie and become that user.
Why it existsOne server can fail and nobody notices. It's also how a system handles a traffic spike without falling over.
Why it existsIt permanently killed "well, it works on my computer."
Why it existsThe same code runs in testing and production with different settings, and no password ever gets committed to version control.
Why it existsThe one category of value that must never reach a user's browser. Treat a leaked key as compromised the moment it's exposed, not when someone abuses it.
Why it existsIt removes the step where a human forgets to run the tests before shipping on a Friday.
Why it existsSo a mistake costs a redeploy instead of a customer. A vendor with no staging environment is testing on you.
Why it existsModern teams deploy many times a day. A shop that deploys quarterly usually has a fear problem, not a quality one.
Why it existsSpeed of undo is a feature. It's why database changes are numbered and every release is versioned.
Fair question for any vendor: "If a release breaks something, how fast can you put it back?"
Why it existsReports, syncs, cleanup. The classic failure is that it stops running and nobody notices for weeks, because success is silent.
If a scheduled job matters, it needs an alert on failure. Silence is not evidence it ran.
Why it exists99.9% sounds identical to 99.99% and is ten times more downtime — about nine hours a year versus fifty minutes.
Why it existsThe first place anyone looks when something breaks. A system with no logs can only be debugged by guessing.
Why it existsThat one mechanic, at enough scale, turns out to cover writing, reasoning, summarizing, coding, and using tools.
Why it existsEvery cost and every limit in AI is measured here. Rough math: 1,000 tokens is about 750 words.
Nothing to do with a login token. Same word, unrelated meaning — see the overloaded words below.
Why it existsIt's the hard ceiling on "just paste the whole manual in." This single constraint is why retrieval systems exist.
Why it existsIt's the entire input, not just the question you typed — which is why prompt design turns out to be most of the work.
Why it existsPersistent behavior without repeating yourself every message. It's how a generic model becomes your company's assistant.
Why it existsDistinguishes using a model — cheap and instant — from training one, which is neither.
Why it existsIt solves three problems at once: your documents don't fit in the context window, sending them all would cost a fortune, and accuracy drops when the answer is buried in noise.
This is what "train the AI on our documents" almost always actually means — and it's far cheaper and faster than real training.
Why it existsIt lets a computer match "how do I reset a customer's password" to a document titled "credential recovery procedure." Keyword search cannot do that.
Why it existsAn ordinary database matches exact values. This one matches meaning. It's the storage half of any RAG system.
Why it existsRetrieve a whole 200-page manual and you're back to the context problem. Chunk size is the main tuning dial in a retrieval build.
Why it existsThe jump from "writes you an answer" to "completes the task." This is what most AI automation projects are actually buying.
Why it existsLets a model act on live company data instead of only on what it memorized during training.
Why it existsBefore it, every model-and-tool pairing needed its own bespoke integration. Now one connector works everywhere. That's the entire value.
Increasingly the answer to "how do we connect our AI to our actual systems?"
Why it existsRarely the right first move. Better prompting and retrieval solve most of what people reach for fine-tuning to fix, at a small fraction of the cost and time.
If a vendor opens with fine-tuning, ask what they tried first. It's often a sign of an expensive answer to a cheap problem.
Why it existsNot a bug on a roadmap — a property of how prediction works. Serious systems are designed around it with source citations, grounding, and human approval steps.
Any vendor promising it's "solved" is overselling. The right answer describes the guardrails, not the absence of the problem.
Why it existsPulling data out of invoices wants low. Writing five ad variations wants high. One setting cannot serve both.
Why it existsThe only way to know a change improved things rather than quietly breaking them. Most teams skip it, which is why their AI features silently degrade.
"How do you measure whether it's getting better?" separates a real AI build from a demo.
Why it existsLets you deploy automation before you fully trust it. The review step is removed later, once the numbers earn it — not on day one.
The real trap
Same word, two meanings
Most confusion in a technical meeting doesn’t come from words you’ve never heard. It comes from words you know that quietly mean something else in the room. Nobody warns you about these.
token
Logins
A signed credential proving who you are.
AI
A chunk of text, about ¾ of a word. The unit models are billed in.
role
Your app
A person's job — admin, manager, viewer.
The database
A database identity with its own permissions. Unrelated to job titles.
client
Business
The company paying the invoice.
Software
The user's browser or device. "Client-side" means on their machine.
key
Security
A secret that grants access — an API key, a signing key.
Database
An identifier for a record. A primary key is just an ID column.
schema
Database
The structure of your tables and the type of each field.
Validation
A rule describing what counts as valid input on a form.
model
AI
The trained system itself — Claude, GPT.
Software
The shape of a thing in your app. "The customer model" means its fields.
function
Code
A named block of code you can run.
AI
A tool the model can invoke — "function calling."
agent
AI
A model with tools, looping toward a goal.
Browsers
"User agent" is just the browser identifying itself. Completely unrelated.
cache
Three of them
The browser's, the delivery network's, and the application's — all separate.
Why it matters
"Stale content" has three possible fixes. Naming which cache is most of the diagnosis.
session
Login
The system's memory that you're currently signed in.
Analytics
One visit to your site, usually ending after 30 minutes idle. Different thing entirely.
context
AI
The total text a model can consider at once.
Business
Background information. When a vendor says "context window," they mean the first one.
migration
Database
A numbered file that changes your data structure.
Infrastructure
Moving an entire system to a new host or platform. Vastly different scope.
The method
How to decode a word that isn’t on this list
Ask what existed before it
Every term here names a solution to a problem someone had. Row-level security exists because developers kept forgetting filters. Retrieval exists because documents don't fit in a prompt. Find the problem and the term explains itself.
Ask where it runs
On the user's device, on the company's server, or inside the database? Three very different levels of trust, and the answer usually explains why the thing is built the way it is.
Ask what it replaced
New tools are almost always old tools with one constraint removed. A connector protocol is an API with a standard shape. An embedding is a search index that understands meaning instead of spelling.
Say "I don't know that one"
Out loud, in the actual meeting. Good engineers read this as confidence and will happily explain. Nodding along is the only version that costs you credibility — and it's how bad scope gets approved.
Knowing the words is step one.
We train teams to use AI on their actual work, and we build the automations that remove the repetitive parts of running a service business. Both start with a conversation, not a contract.