A fun, no-shame learning tool for non-techies who need to understand the language of AI, automations, apps, data, and cloud systems while they work in fast-moving technology industries.
Public — no login needed Online learning app Progress saves on this device
Switch tracks any time. Your XP stays with you in this browser.
Software that can do tasks that usually need human judgment, like writing, sorting, predicting, or recognizing patterns.
Pick three terms from today's meeting and rewrite them in customer-friendly language.
+25 XPFind one place an AI or automation should stop and ask a human before acting.
+30 XPTake one intimidating term and explain it using a restaurant, sports, or road-trip analogy.
+20 XPSoftware that can do tasks that usually need human judgment, like writing, sorting, predicting, or recognizing patterns.
A very fast intern that learned from a huge library, but still needs clear direction and review.
Use AI to draft, summarize, classify, or brainstorm — then verify anything important.
AI is not automatically correct, fair, private, or approved just because it sounds confident.
A repeatable set of steps a computer follows to make a decision or produce an output.
A recipe: same ingredients and steps should produce the same kind of result.
When a system ranks leads or recommends a next action, an algorithm is usually behind it.
If the input data is bad, the algorithm can produce bad answers very consistently.
A structured way for one app to ask another app for data or to trigger an action.
A restaurant menu for software: it lists what you can order and how to ask for it.
APIs let tools like CRM, payment, scheduling, and AI systems talk to each other automatically.
APIs usually need permissions, rate limits, and error handling — they are not magic pipes.
A system that performs a repetitive task with little or no manual effort after it is set up.
A conveyor belt for routine work.
Good automations handle reminders, imports, status checks, routing, and reporting.
Automating a broken process usually makes the broken part happen faster.
Computers, storage, and services rented over the internet instead of owned in your office.
Renting a professional kitchen instead of building one in your garage.
Cloud platforms host websites, databases, AI services, backups, and background jobs.
Cloud still has costs, security settings, outages, and access controls to manage.
An organized place where an app stores and retrieves information.
A highly structured filing cabinet that software can search quickly.
Customers, orders, lessons, progress, invoices, and messages often live in databases.
A spreadsheet can hold data, but a database usually gives stronger rules, access, and history.
A chain of steps that moves, cleans, transforms, or analyzes data from one place to another.
A factory line for information.
Pipelines power dashboards, reports, lead scoring, billing exports, and AI training data prep.
When a dashboard looks stale, the pipeline may be stuck even if the app itself is online.
A list of numbers that represents the meaning of text, images, or other content so software can compare it.
A map coordinate for meaning instead of geography.
Embeddings help AI search by meaning, find similar documents, match photos, or group feedback.
Embeddings can find things that are similar, but similarity is not the same as truth.
AI that creates new content such as text, images, code, audio, video, or plans.
A remix machine trained on patterns from many examples.
Use it for drafts, variations, explanations, roleplay, summaries, and quick prototypes.
Generated content can include mistakes, copyright concerns, or confidential information if used carelessly.
When AI confidently gives information that is wrong, made up, or not supported by the source material.
A person guessing with a straight face.
Watch for fake citations, incorrect policies, bad calculations, or invented customer details.
The more specific an unsupported answer sounds, the more tempting it is to trust it.
Large Language Model: an AI model trained to understand and generate language.
Autocomplete with a giant memory of language patterns and reasoning examples.
ChatGPT, Claude, Gemini, and many workplace AI assistants are powered by LLMs.
An LLM predicts useful responses; it does not inherently know your company policy or live facts.
A way for software to learn patterns from examples instead of only following hand-written rules.
Training by examples, like showing many photos until someone learns what a good one looks like.
ML can predict churn, classify tickets, detect fraud, recommend next actions, or score images.
The model learns from the examples it gets, including their bias and gaps.
The trained system that turns inputs into predictions, classifications, or generated output.
The brain inside the tool, trained for a type of task.
Different models are better for writing, coding, vision, speed, privacy, cost, or accuracy.
Newer or bigger does not always mean better for your workflow.
The instructions, question, examples, or context you give an AI tool.
A work order for an AI assistant.
Better prompts include the role, goal, context, constraints, examples, and desired output format.
A vague prompt usually produces a vague answer.
Retrieval-Augmented Generation: AI answers using retrieved source documents or data instead of only memory.
An assistant that checks the file cabinet before answering.
RAG is useful for company knowledge bases, policy bots, support assistants, and document-heavy work.
RAG is only as good as the sources it retrieves and the permission rules around them.
A small chunk of text an AI model reads or writes. AI costs and context limits are often measured in tokens.
Word pieces on an AI meter.
Long documents, chat history, and repeated instructions use more tokens and can cost more.
Token limits mean an AI may not remember everything in a long conversation or document set.
An AI-powered worker that can plan steps, use tools, inspect results, and keep going toward a goal.
A digital teammate with a checklist and access to approved tools.
Agents can research, update records, run checks, draft reports, or monitor queues under guardrails.
Agents need clear permissions, logging, review points, and stop conditions.
Rules, checks, and limits that keep a system from doing unsafe, unauthorized, or low-quality things.
Lane lines and speed limits for automation.
Guardrails can block live sends, require approval, redact secrets, validate output, or stop risky actions.
A guardrail that is only written in a policy doc will not stop a bad automation by itself.
Minimum Viable Product: the smallest useful version that proves whether the idea solves a real problem.
A food-truck version of a restaurant concept.
An MVP should teach you whether users care before you build every feature.
Minimum does not mean sloppy; it must still solve the core job reliably.
Tools that let people build workflows, forms, apps, or automations with little traditional programming.
Building with labeled blocks instead of cutting every piece of wood yourself.
Good for prototypes, internal tools, automations, and admin workflows.
No-code can still create fragile processes, hidden costs, and security risks if nobody owns it.
The delay between asking a system to do something and getting a response.
The wait time after you press a button.
High latency can make apps feel broken, even when they technically work.
AI, databases, APIs, and poor network paths can all add latency.
A database designed to store embeddings and find items that are similar in meaning.
A library organized by ideas instead of alphabetically.
Vector databases power semantic search, recommendations, document assistants, and RAG systems.
It does not replace source-of-truth databases; it helps find relevant context.
An automatic message one system sends to another when something happens.
A doorbell for software events.
A payment app can send a webhook when an invoice is paid, triggering a receipt or task update.
Webhook receivers need verification, retries, and duplicate handling.
0 correct so far
Your team wants AI to answer questions from company policies and cite the policy page.
RAG means the AI retrieves source documents before generating an answer.
A payment system should notify your dashboard the moment an invoice is paid.
A webhook is an event message from one system to another.
An AI answer sounds confident, but nobody can find the source and the date is wrong.
A hallucination is unsupported or made-up output presented confidently.
The app works, but users complain every button feels slow after they click.
Latency is the time between request and response.
You need the smallest useful version of a product to test whether customers care.
An MVP is the smallest version that can validate the core value.