LLMs. Agents. MCP. RAG.What do these terms actually mean?
You've heard all four this month alone — in a job posting, a LinkedIn post, a meeting you nodded through. This issue is the plain-English version nobody gave you the first time.
Why nowThese 5 words show up in nearly every AI job posting
If you read nothing else
An LLM is the engine. An Agent is that engine given a task and the ability to act on it. MCP is how it plugs into your tools and data. RAG is how it looks things up instead of guessing. A context window is how much it can hold in its head at once. That's the whole map.
🎙️ Nova's Signal
Nova is FOWL AI's news anchor — she tracks the signals every week.
"Every one of these terms sounds harder than it is. That's not an accident — vague jargon makes a feature sound more impressive than a plain sentence would. Once you can define all five in under ten seconds each, most AI product pitches stop sounding magical and start sounding like what they are: useful, specific tools."
Nova's call: Bookmark this one. You'll reference it more than any issue we've done.
New here? — This is a Reference Issue, same format as Issue 4 (The AI Model Map). No prior AI knowledge assumed — every term is defined from zero before we go further.
Somewhere in the last few months, five words started showing up everywhere — job postings, product launches, LinkedIn posts, meetings — and most people nodded along without a clean definition for any of them.
That's not a knowledge gap you caused. Almost nobody explains these terms in order, from the ground up, without assuming you already know the last one. So they stack: you half-understand "LLM," which makes "Agent" fuzzy, which makes "MCP" sound like alphabet soup.
This issue fixes that. Five terms, each defined in one plain sentence first, then unpacked with a real example. Read it once and you'll have a permanent answer key.
5 Terms, Plain English
01
LLM — Large Language Model
One sentence: an LLM is software trained on huge amounts of text that predicts what word should come next, which turns out to be enough to hold conversations, write code, and answer questions.
That's it — underneath everything, an LLM is a very sophisticated next-word predictor. It doesn't "know" facts the way a database does; it's learned patterns from the text it was trained on and generates a response one piece at a time, each piece chosen based on everything that came before it. ChatGPT, Claude, and Gemini are all products built around an LLM — the LLM is the engine, the product is the car built around it.
Why it matters · Every other term in this issue is something built on top of an LLM. This is the foundation the rest of the map sits on.
The signal — When someone says "the model," they usually mean the LLM specifically — not the whole product wrapped around it.
02
Agent — an LLM given a task and the ability to act
One sentence: an AI Agent is an LLM that's been given a goal, a set of tools it's allowed to use, and the ability to keep working in steps until the goal is done — instead of just answering one question and stopping.
A plain chatbot answers what you type and waits. An agent can be told "book me the cheapest flight to Chicago next Tuesday," then actually search flights, compare prices, and complete the booking — checking its own work and adjusting along the way, without you approving every single step. The shift from "answers questions" to "completes tasks" is the entire reason "AI Agent Orchestration Specialist" showed up as one of the most in-demand new job titles this year.
Why it matters · Nearly every "next generation of AI" headline you've seen this year is really just describing agents getting more reliable at multi-step tasks.
The signal — If a tool can only respond to one message at a time with no memory of what it just did, it's not an agent yet — it's a chatbot.
03
MCP — Model Context Protocol
One sentence: MCP is a standard way for an AI model to connect to outside tools and data — your files, a database, a calendar, a company's internal system — without every company having to build a custom connector from scratch.
Before MCP, if you wanted an AI to read your Google Calendar, someone had to build a one-off integration just for that. MCP (created by Anthropic, now adopted industry-wide) works more like a universal power outlet: any tool that speaks MCP can plug into any AI model that supports it, and they understand each other automatically. It's plumbing, not a feature you'd notice directly — but it's the reason AI agents can now reach into real software instead of being stuck in a chat window.
Why it matters · MCP is what turns an agent from "smart but isolated" into "smart and actually connected to your work."
The signal — If you see a company advertise "MCP support," they mean their product can be plugged into by AI tools — that's a compatibility claim, not a product feature on its own.
04
RAG — Retrieval-Augmented Generation
One sentence: RAG is a technique where an AI looks up relevant information from a specific set of documents before answering, instead of relying only on what it memorized during training.
An LLM's training data has a cutoff date and doesn't include your company's internal wiki, your specific contracts, or last week's news. RAG fixes that by searching a document store for the most relevant pieces of text, then handing those to the model along with your question — so the answer is grounded in something real and current, not just the model's memory. This is why an AI customer support tool can answer questions about your specific product instead of hallucinating a generic answer.
Why it matters · RAG is the main defense against an AI confidently making things up — it's the difference between "guessing" and "looking it up."
The signal — If an AI tool cites a specific source or document for its answer, RAG is almost certainly running underneath it.
05
Context Window — how much the model can hold in mind at once
One sentence: the context window is the total amount of text — your messages, its replies, any documents you've shared — that a model can "see" and reason about at one time, measured in chunks of text called tokens.
Once a conversation grows past the context window's limit, the oldest parts start dropping out of view — which is why an AI can seem to "forget" something you told it twenty minutes ago. Bigger context windows (some now stretch past a million tokens) let a model hold an entire codebase or book in mind at once, which is a big part of why AI tools have gotten noticeably better at long, complex tasks over the past year.
Why it matters · When an AI response gets worse in a long conversation, a full context window is often the real culprit — not a sudden drop in the model's ability.
The signal — "Tokens" are roughly ¾ of a word on average — so a 1-million-token context window is very roughly 750,000 words, or several long novels.
📡 Signal & Chatter
What the data and community are saying this week.
›
"Agent" is the most overused word in AI marketing right now. A growing share of products labeled "AI Agents" are really just chatbots with a new coat of paint — ask whether it can complete a multi-step task unsupervised before you believe the label.
›
MCP adoption has moved fast. Since its release, major AI products and developer tools have added support for it, which is unusually quick for an open standard — a sign the industry wanted this problem solved.
›
RAG is quietly becoming table stakes. Most serious enterprise AI tools now use it by default — a model with no retrieval layer is increasingly treated as a demo, not a product.
›
Context windows keep growing. What counted as a large context window two years ago is now considered small — treat any specific number you read as a snapshot, not a ceiling.
🔮 FOWL Prediction #9
"By Issue 20 (early 2027), 'Can you use MCP?' will show up in AI-related job interviews as often as 'Do you know Git?' shows up in engineering interviews today."
MCP adoption is moving faster than most open standards do, and it's becoming the default way AI tools connect to real work. We'll check back on this prediction when Issue 20 goes out.
FOWL AI · July 20, 2026 · We'll score this in early 2027.
✅ 3 Things to Do This Week
One for each type of reader — pick yours
01
If you're job hunting in or near AI — re-read job postings you've saved and circle every instance of these 5 words. You'll immediately understand what the role actually involves, not just the buzzword version.
02
If you use AI tools day to day — next time a long conversation starts giving worse answers, start a fresh one instead of assuming the model got dumber. It's very likely the context window, not the model.
03
If you evaluate or buy AI products — ask any vendor claiming to have an "AI Agent" one direct question: what multi-step task can it complete without you approving each step? The answer will tell you fast whether it's a real agent or a chatbot with better marketing.
🚀 From the Lab
Sponsored by FOWL AI
Launch Lab — Building something?
We're not an agency — we don't build it for you. Brand Brain is the AI workspace you use to turn what you know into a website, newsletter, and digital products you actually own. You stay in control of it the whole way.
This title barely existed two years ago. Now it's one of the fastest-growing roles tied directly to term #2 in this issue: someone who designs, connects, and supervises multiple AI agents working together on a business process — routing tasks between them, setting the guardrails, and catching the failures before they become a customer-facing problem.
It's less about writing code and more about systems thinking — understanding a workflow well enough to break it into steps an agent can reliably execute, and knowing exactly where a human still needs to check the work. People moving into this role come from operations, project management, and QA backgrounds as often as from engineering.
How to position for this now: If you've ever mapped out a multi-step business process — even in a spreadsheet — that experience is more transferable to this role than most people realize. Lead with the process-design skill, not a coding background you don't have.
None of these five terms are actually complicated once someone hands you the plain-English version first. The complexity was never in the concepts — it was in nobody bothering to define them in order.
Next time one of these words comes up in a meeting, a job posting, or a product pitch, you'll know exactly what's being claimed — and just as importantly, what isn't.
"Jargon isn't complexity. It's just a definition nobody gave you yet."
💬 One question — reply and tell us
Which AI term do you keep hearing but still can't define confidently? Hit reply — we'll cover it in a future issue.
Hit reply. We read every one and it shapes what we cover next.