Quick takeaways

  • Use the glossary to align terms before debating tools or architecture.
  • Definitions here are practical, not academic.
  • When in doubt, tie the term back to a real workflow or decision.

AI terms and concepts explained

Gaurav Sen explains 20 core AI concepts in 40 minutes — tokenization, prompting, RAG, vector databases, MCP, and agents — in the same plain-English style as this glossary.

Glossary terms

The terms below are grouped by how deep you need to go to use them. Beginner terms are useful in almost every AI conversation. Intermediate terms matter once you start building workflows. Advanced terms become relevant when you design RAG, agents, or evaluation systems.

Beginner terms

PromptThe instruction or task you give the model. Example: "Summarize this transcript into three decisions and two open questions."
TokenA piece of text the model processes. Tokens affect cost and context size. Longer inputs and outputs cost more.
Context windowHow much text the model can consider at once. Large documents may need chunking or summarization.
HallucinationA confident but incorrect AI output. Always check facts before acting on them.
Review gateA human approval point before the workflow continues. Example: a founder reviews an AI-drafted investor update before sending.

Intermediate terms

EmbeddingA numerical representation used to compare meaning across text. Useful for search, clustering, and retrieval.
RAGRetrieval-augmented generation: retrieve relevant source material before answering. Read the RAG guide.
MCPModel Context Protocol: a structured way for AI to connect to tools and context. Read the MCP guide.
AgentAn AI workflow that can plan, use tools, and iterate toward a goal. Agents still need review gates and logging.
EvalA test or rubric used to measure output quality. Example: "Does the answer match the source document?"
ChunkingSplitting documents into smaller pieces so they fit into context and retrieval systems. Good chunks preserve meaning.
TemperatureA setting that controls randomness. Lower values produce more consistent outputs; higher values produce more varied outputs.

Advanced terms

Vector databaseA store for embeddings that supports similarity search. Often used in RAG pipelines.
Retrieval evaluationMeasuring whether the right chunks were found for a question. Often scored with recall, precision, or MRR.
Tool callingWhen a model decides to invoke an external function or API instead of answering directly.
System promptBackground instructions given to the model, separate from the user's actual question. Used to set role, format, and rules.
Fine-tuningTraining an existing model further on your own data. Useful for specialized style or format, but rarely the first step.
MultimodalModels that can process more than text, such as images, audio, or video.
Reasoning modelA model trained or prompted to show intermediate steps before producing a final answer.
LatencyThe delay between sending a request and receiving a response. Important for real-time or high-volume workflows.

How to use this glossary

Use the glossary before projects, vendor reviews, and team training. If a discussion is getting abstract, point back to the workflow: what is the input, what is the expected output, and where does human review happen?

  1. When you hear an unfamiliar term, look it up here first.
  2. Ask how the term connects to your workflow before investing in it.
  3. Share the definition with teammates to keep conversations aligned.
  4. Update the glossary when your team adopts new patterns or tools.
Related: Read the AI roadmap for the learning order and the adoption checklist for rollout language.

Without AI vs. with AI

TaskWithout AIWith AI
Term alignmentTeams argue over words like agent, RAG, and eval without shared definitions.The glossary gives plain-English definitions tied to real workflows so conversations start aligned.
OnboardingNew hires search the web and get inconsistent or academic explanations.The glossary provides practical definitions written for founders, developers, and marketers.
Vendor discussionsSales teams use jargon that technical reviewers interpret differently.The glossary lets teams point to a shared definition before debating tools or architecture.
Decision clarityLeaders approve projects without understanding model limits or review gates.The glossary defines concepts like hallucination, context window, and review gate in decision terms.
Cross-functional workMarketing, product, and engineering use the same word to mean different things.The glossary creates a shared reference that reduces miscommunication across teams.

FAQ

Is this glossary technical?

Only as much as it needs to be. The focus is on practical understanding for real work.

What should I read after the glossary?

Start with prompt engineering, then move to RAG, MCP, and agents depending on your workflow.

Do I need to know every term?

No. Learn the terms that show up in the workflows you are actually building.

How is this different from other AI glossaries?

These definitions tie each term to a workflow decision, not just a theory concept.

Can I share this with my team?

Yes. Use it as a shared reference before architecture reviews or vendor calls.

Should beginners start with advanced terms?

No. Start with beginner terms and only move deeper when a workflow actually requires it.

Is this AI glossary technical?

Only as much as needed for real decisions. The focus is practical understanding, not academic depth.

What should I learn after the glossary?

Start with prompt engineering, then move to RAG, MCP, agents, and evaluation based on your workflow.

Do I need to memorize every term?

No. Learn the terms that appear in the workflows you are actually building.

How is this glossary different from others?

Definitions are tied to workflows, decisions, and review gates rather than pure theory.

Beginner

How to learn AI in 2026

A practical learning path from AI literacy to agentic workflows.

Read guide
RAG

What is RAG? Retrieval-augmented generation explained

A practical guide to RAG for teams that need AI answers grounded in documents, customer data, policies, knowledge bases, or frequently changing information.

Read guide
MCP

What is MCP? Model Context Protocol explained

Learn how MCP helps AI assistants connect to tools, files, APIs, and business systems so they can do useful work with the right context.

Read guide