From AI to Agentic AI: The 7 Terms Everyone Should Be Able to Explain
Seven terms that get thrown around as if they mean the same thing. They do not. Work through the diagram, examples, two hands-on labs, three exercises and a quiz, and you will be able to explain each one in a sentence.
The key idea in one line
These concepts are related, not interchangeable. Think of them as a hierarchy that keeps building up: each one is a more specific kind of the one before it, and the last two are systems built on top of LLMs.
Why this matters
- Understand the hierarchy and how the pieces relate
- Choose the right approach for your problem (you do not need an agent to detect spam)
- Use RAG and agents the right way for accuracy and autonomy
Part 1
The hierarchy
Everything starts here. AI is the outer circle. Each inner circle is a narrower, more specific technique. RAG and Agentic AI are not circles inside the others: they are systems that take an LLM and wrap extra parts around it.
Click any ring or box.
Artificial Intelligence
The broad umbrella
Machines that think, reason and act intelligently.
Any software that does something we would call "intelligent" if a person did it: reasoning, planning, understanding language, recognising objects, or making decisions. It does not have to learn. A program that follows thousands of hand-written rules still counts as AI.
Part 2
Each concept, explained simply
For every term: what it is, an everyday analogy, a concrete example, the products you have heard of, and where it shows up in Appian. Go through all seven.
1/7 explored
Artificial Intelligence
The broad umbrella
What it is
Any software that does something we would call "intelligent" if a person did it: reasoning, planning, understanding language, recognising objects, or making decisions. It does not have to learn. A program that follows thousands of hand-written rules still counts as AI.
Think of it like
AI is like the word "vehicle". Cars, bikes and buses are all vehicles. Everything else on this page is a type of AI, the same way a car is a type of vehicle.
A concrete example
A chess program from the 1990s beat grandmasters using rules and brute-force search written by humans. Almost no learning was involved, yet it is clearly AI.
Real-world examples
Part 3
How we got here
Each step exists because the previous one hit a limit. Walk the chain and the whole picture clicks into place.
Step 1 of 7
Artificial Intelligence
We start by wanting machines that can think, reason and act.
Machines that think, reason and act intelligently.
Lab 1
RAG lab: closed book vs open book
Ask the same question to a plain LLM and to an LLM with retrieval. Watch what the retrieval step adds.
1. Pick a question
2. Choose the setup
Generating…
Lab 2
Agent loop: watch an agent work
Press play and follow one goal through the plan, tool, observe, reason, act loop that makes an agent different from a chatbot.
1. Goal
"Book me the cheapest flight to Delhi next Friday and add it to my calendar."
Part 4
Key modern terms to know
Eight terms you will hear in every AI conversation this year, including two you will meet the first time you configure a Generative AI Skill. Tap a card for a concrete example.
Exercise
Sort it: which layer is this?
Eight real-world situations. Pick the most specific term that describes each one. You get instant feedback and the reason.
1A property site estimates a house price from size, location and past sales
2Your phone unlocks by recognising your face
3A tool writes a brand-new song from a text description
4A chatbot answers questions using only your company's policy documents
5An assistant plans a trip, books the hotel and emails the itinerary
6A thermostat with hand-written rules: "if below 18°C, turn on heating"
7You ask a model to explain quantum physics and it writes three paragraphs
8A bank flags a card payment as fraud based on past transactions
Exercise 2
Appian drill: which concept is this feature?
Five things you can build in Appian today. Name the concept behind each one. This is the mapping an interviewer will ask you for.
1You upload 30 sample invoices and 30 sample contracts, and the Classify Documents skill learns to tell them apart
2A Generative AI Skill turns a long complaint email into a three-line summary
3An Agent Studio agent is given a tool that queries the Cases record type, and answers "what is the status of case 4471?" from that data
4An agent reads a new ticket, decides it is a bug, creates a Jira issue through an integration, and starts an Appian process to notify the owner
5A decision object with hand-written rules routes a request to Legal when the amount is over 50,000
Exercise 3
Prompt, fine-tune or RAG?
Four situations from real projects. Pick the cheapest technique that actually solves the problem.
1Answers must quote the current version of your HR policy, which changes every quarter
2The summaries are fine but too long and too formal for the sales team
3Every output must follow your firm's exact contract drafting style, learned from 2,000 past contracts
4The agent keeps answering questions about customer orders it cannot possibly know
Final check
Twelve-question quiz
Score 10 or more and you can confidently explain all seven concepts, and the terms around them, to someone else.
Which statement about the hierarchy is correct?
Take-away
One-line cheat sheet
Bookmark this. If you can say each line from memory, you are done.
| Term | In one line | Example | In Appian |
|---|---|---|---|
| AI | Machines that think, reason and act intelligently. | Expert systems | Every AI feature, from AI Skills to Agent Studio |
| ML | Machines that learn patterns from data. | Spam detection | Trained Classify Documents skill |
| DL | ML that learns complex patterns with many-layered neural networks. | Image recognition | Document extraction skills |
| GenAI | Machines that create new content. | ChatGPT | Generative AI Skills (generate, summarise, extract) |
| LLMs | Machines that understand and generate language. | GPT-4 / GPT-5 | The model behind those skills: GPT, Claude, Nova |
| RAG | Machines that combine retrieved knowledge with GenAI. | Enterprise chatbots | An agent tool that looks up records or documents first |
| Agentic AI | Machines that act autonomously to reach goals. | OpenAI Agents | Agent Studio, run by the Execute AI Agent smart service |
Sources
References: check it yourself
Every Appian statement above comes from the 26.8 documentation, and every concept from the paper or primer that defined it. Each link says which claim it backs.
Appian (26.8 documentation)
- What are AI Skills? Appian Docs
The skill types: classification, extraction, generation, summarisation, PII detection; AI Guardrails on every generative execution.
- Generative AI Skills Appian Docs
The model choices (Appian-hosted GPT, Claude and Nova via AWS, Azure OpenAI, custom), the Auto option, AI Actions, and the Execute Generative AI Skill smart service.
- Agent Studio Appian Docs
An agent as a natural-language prompt plus reusable tools into the data fabric and processes, run by the Execute AI Agent smart service.
The concepts
- Machine Learning Glossary Google for Developers
Working definitions of ML, deep learning, embeddings, tokens and most terms on this page.
- What is machine learning? IBM
ML as learning patterns from data instead of hand-written rules, and where it sits inside AI.
- Deep Learning (Goodfellow, Bengio, Courville) MIT Press, free online
Why stacking layers lets a network learn edges, then shapes, then objects. The standard textbook.
- Deep Blue IBM History
The 1997 chess example: search and hand-tuned evaluation, almost no learning, still AI.
- Attention Is All You Need Vaswani et al., 2017 (arXiv)
The Transformer, the architecture behind every LLM named on this page.
- Embeddings guide OpenAI Platform docs
Text as number lists that capture meaning, and similarity search over them.
RAG, hallucination and agents
- Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks Lewis et al., 2020 (arXiv)
The paper that named RAG: retrieve passages first, then generate from them.
- A Survey on Hallucination in Large Language Models Huang et al., 2023 (arXiv)
What hallucination is, why next-word prediction produces it, and why grounding reduces it.
- ReAct: Synergizing Reasoning and Acting in Language Models Yao et al., 2022 (arXiv)
The reason, act, observe loop the agent lab animates.
- Building effective agents Anthropic
When a plain LLM call is enough and when a tool-using loop is worth it; the advice behind "you do not need an agent to detect spam".
Appian documentation links point at version 26.8. Appian moves pages between releases; if one is gone, search the docs for the title.