One of the hardest things that make any subject you’re trying to learn difficult is the vocabulary. When you’re trying to learn (or do) something that is laced with words you don’t understand, it will be almost impossible.
Not only that, but people tend to want to avoid topics they don’t understand. And, frankly, AI is just too important for people to avoid.
A lot of what makes AI feel intimidating isn’t the technology itself. It’s the vocabulary.
You read an article or watch a video, and within two minutes someone’s talking about tokens, context windows, MCP servers, agents and RAG. Everybody nods along like this is common knowledge. And if you’re new to it, you walk away feeling like you missed a class somewhere.
You didn’t. The AI world just invented (or borrowed) a big pile of words in a very short amount of time. Some of them are genuinely useful to understand. Some of them are marketing. And a few of them mean completely different things depending on which company is using them.
In the first post in this series, I covered what AI actually is and isn’t. This one is the dictionary that goes with it. I’m not going to give you one-line definitions that leave you just as confused as before. For each term, I’ll explain what it means in plain English, and why it matters to someone running a small online business. (If it doesn’t matter much, I’ll tell you that too.)
You don’t need to read this top to bottom. The terms are grouped by topic, and there’s a quick index right below so you can jump straight to whatever term has been bugging you. And bookmark it… I’ll keep it updated, because this vocabulary is changing fast.
Let’s dive in.
The Basics
These are the foundational terms. If you get these straight, a lot of the rest falls into place.
AI (Artificial Intelligence)
Technically, “AI” is a huge umbrella term that’s been around since the 1950s. It covers everything from the spam filter in your email to the algorithm that decides what shows up in your Facebook feed.
But when people say “AI” today, they almost always mean one specific kind: tools like ChatGPT, Claude and Gemini that you can talk to in plain English and that write, summarize, research and increasingly do things for you.
So when you see “AI” in this series, that’s what I mean. Just know that the word gets slapped on all kinds of products, some of which have very little to do with ChatGPT-style AI. “AI-powered” on a product label doesn’t tell you much by itself.
Generative AI
This is the more precise name for the current wave. “Generative” means it generates new stuff… text, images, audio, video, code… rather than just sorting or analyzing existing data.
Your email’s spam filter is AI, but it’s not generative. It just decides “spam or not spam.” ChatGPT is generative. You ask it for a product description and it writes one that didn’t exist a second ago.
You’ll see “GenAI” as shorthand for this too.
LLM (Large Language Model)
This is the engine under the hood of ChatGPT, Claude, Gemini and most of the other text-based AI tools.
An LLM is a system that was trained on an enormous amount of text… a big chunk of the internet, books, articles, code… until it got extremely good at understanding and producing language. “Large” refers to the sheer size of it, both in how much it was trained on and how big the model itself is.
I explained how an LLM actually works in the first post, but the short version is: it learned by playing a guess-the-next-word game trillions of times, and in the process it picked up patterns in how language and ideas work.
When people talk about “the AI,” they’re usually talking about an LLM.
Model
A model is a specific version of an AI engine. GPT-5 is a model. Claude Opus and Claude Sonnet are models. Gemini Pro and Gemini Flash are models.
Here’s the thing that trips people up: the model and the app are not the same thing. ChatGPT is an app. Inside it, you’re using one of OpenAI’s models. Claude.ai is an app that runs Anthropic’s Claude models. And companies release new models all the time, often several sizes at once. A bigger model is usually smarter but slower and more expensive. A smaller model is faster and cheaper, and plenty good for simple jobs.
Why this matters to you: when someone says “ChatGPT got way better,” what usually happened is that OpenAI swapped in a newer model. And when you pay for a higher-tier plan, a big part of what you’re buying is access to the best models.
Frontier Model
A frontier model is one of the most advanced AI models available at any given moment… the ones at the “frontier” of what AI can currently do. You’ll also hear the companies that build them called frontier labs. Right now that mainly means OpenAI, Anthropic and Google, with a few others like xAI close behind.
Frontier models are the flagships. They’re the smartest, the best at complex reasoning and writing, and usually the most expensive to use. Below them are smaller, faster, cheaper models, older models, and open models you can download and run yourself (which tend to trail the frontier by a few months).
Two things worth knowing. First, the frontier moves fast. Today’s frontier model is next year’s budget option. Something that felt like magic a year ago is now what the free plans give you. Second, you don’t always need the frontier. For quick jobs like rewording an email or summarizing an article, a smaller model does fine and is faster. Save the top model for the work that actually needs the horsepower: strategy, tricky technical problems, and anything where quality really matters.
Chatbot (or AI Assistant)
The app you actually type into. ChatGPT, Claude, Gemini, Copilot, Grok… these are all AI assistants, each running its company’s models behind the scenes.
“Chatbot” is a slightly dated term now, because these tools do a lot more than chat. But you’ll still hear it a lot, and it’s also used for the little chat windows on business websites (which may or may not be running a real LLM).
Prompt
A prompt is whatever you type (or say) to the AI. A question, an instruction, a pasted document with “summarize this”… it’s all a prompt.
This matters more than it sounds like it should. The quality of what you get out depends heavily on what you put in. A vague one-line prompt gets you a vague, generic answer. A prompt that gives context (“I run a membership site for knitters, here’s my last three emails, write the next one in the same tone”) gets you something useful.
I’ll have a whole post later in this series on writing better prompts. For now, just know that “prompt” simply means your side of the conversation.
Multimodal
A multimodal AI can work with more than just text. It can look at images, read PDFs, listen to audio, and in some cases watch video or generate images and speech.
In practice, this is really handy. You can screenshot an error message on your WordPress site and ask “what does this mean?” You can upload a photo of a whiteboard and have it turn your scribbles into a to-do list. You can hand it a PDF invoice and have it pull out the numbers.
All the major AI assistants are multimodal now, at least to some degree.
How It Thinks (and Forgets)
These terms explain a lot of AI’s quirks… why it sometimes forgets things, why it’s confidently wrong, and why it doesn’t know about last week’s news.
Training Data
The mountain of text (and images, code, etc.) that a model learned from before it was released. This is where all of its “knowledge” comes from.
Two things worth knowing. First, the model doesn’t store its training data like a library of documents it can look things up in. It learned patterns from it, the way you learned English grammar without memorizing every sentence you ever read. Second, whatever was in the training data shapes what the model knows and how it responds. If a topic was barely covered, the model will be weak on it.
Training Cutoff (or Knowledge Cutoff)
The date the model’s training data stops. Anything that happened after that date, the model simply doesn’t know about… unless it can search the web.
Cutoffs are typically months behind the release date, sometimes a year or more. So if you ask about a plugin update from last month or a recent Google algorithm change, a model without web search might give you confidently outdated information.
Most AI assistants can now search the web, which helps a lot. But it’s worth knowing the cutoff exists, especially for anything time-sensitive. When in doubt, ask it to search.
Parameters
You’ll see model sizes described in parameters: “70 billion parameters” and so on. Parameters are the internal numeric settings that got adjusted during training. They’re where the model’s learned patterns live.
More parameters generally means a more capable model (and a more expensive one to run). But honestly? This is one of those terms you can safely ignore. You’ll rarely need it, and the big companies often don’t even publish the numbers anymore.
Token
A token is the unit AI uses to read and write text. It’s usually a chunk of a word. Short common words are often one token. Longer words get split into a few.
A good rule of thumb: 1,000 tokens is roughly 750 words of English.
Why should you care? Two reasons. First, tokens are how AI usage gets measured and priced. If you ever use an AI through an API (more on that below), you pay per token, both for what you send in and what comes back. Second, tokens are how the size of the context window is measured… which is the next term, and it’s an important one.
Context Window
The context window is how much the AI can “hold in its head” at one time. It’s everything in the current conversation: your messages, its replies, any documents you uploaded, and its instructions, all measured in tokens.
This explains one of the most common frustrations people have. You’re deep into a long conversation, and the AI starts forgetting things you told it at the beginning. That’s usually the context window at work. Once a conversation gets too long, the oldest material gets dropped or summarized to make room.
The headline numbers have gotten huge. As of this writing, the top models from OpenAI, Anthropic and Google all advertise around a million tokens… roughly 750,000 words. But there’s a catch: the chat apps, especially on free plans, often give you a much smaller window than those headline numbers. And even with a big window, AI tends to pay more attention to the beginning and end of a long conversation than the middle.
The practical takeaway: start a fresh conversation for each new task. Don’t run your whole business out of one endless chat thread. You’ll get better results.
Hallucination
When AI makes something up and presents it as fact. Fake statistics, quotes nobody said, citations to articles that don’t exist, a WordPress setting that isn’t real.
It happens because the model is generating the most plausible-sounding answer from patterns, not looking facts up. Usually plausible and true line up. Sometimes they don’t, and the AI delivers the wrong answer in exactly the same confident tone as the right one.
It’s gotten much less common, and web search helps a lot. But it hasn’t gone away. My rule: anything that matters gets verified. Ask for sources, and actually click them.
Reasoning Models (“Thinking” Mode)
A newer kind of model that works through a problem step by step before answering, instead of answering right away. You’ll often see a “thinking” indicator, or be able to expand a section showing its reasoning.
Reasoning models are noticeably better at anything that takes real thought: math, logic, planning, debugging a tricky technical problem, analyzing a complicated situation. They’re slower, and on paid APIs they cost more, because all that thinking uses tokens.
For quick stuff (“rewrite this email to sound friendlier”), you don’t need it. For “help me figure out why my checkout page is broken” or “poke holes in my pricing plan,” turn it on. Most AI assistants now let you choose, or decide automatically.
Thinking Level (Reasoning Effort)
Once you turn thinking on, a lot of tools let you decide how much the model thinks. That’s the thinking level. You’ll usually see it as low, medium and high, and some tools go further with options like “extended” or “max.”
Think of it as a dial. Turn it up, and the model spends longer working through the problem before it answers… checking its own reasoning, considering other approaches, catching its mistakes. Turn it down, and it answers faster with less deliberation.
The names vary from tool to tool, which doesn’t help. You’ll see it called reasoning effort, thinking level, extended thinking, or a thinking budget. In some apps it’s simply a choice between a “fast” version of the model and a “thinking” version. Same idea every time.
Here’s how I’d use it:
- Low for quick stuff. Rewording an email, summarizing an article, simple questions.
- Medium as your everyday default.
- High for work where a wrong answer is costly. Debugging a site problem, planning a launch, analyzing your numbers, poking holes in a strategy.
Higher isn’t automatically better, though. More thinking means more waiting, and it burns through your plan’s usage limits faster (or costs more, if you’re paying through an API). On an easy task, cranking it up to high mostly just makes you wait longer for the same answer. Some tools will pick a level for you based on the question, and that’s usually fine.
Grounding (and Citations)
Grounding means tying the AI’s answer to a real source… a web search, a document you uploaded, your company’s knowledge base… instead of relying only on what it learned in training.
When an AI answer comes with little numbered citations you can click, that’s grounding at work. It’s one of the best defenses against hallucination, because you can check where the claim came from.
When accuracy matters, ask the AI to search and cite its sources. It’s a simple habit that makes a big difference.
Making It Yours
Out of the box, AI knows nothing about you or your business. These are the features and techniques for changing that. (I’ll go deeper on several of these in the next post in the series.)
System Prompt and Custom Instructions
A system prompt is a set of background instructions the AI follows in every conversation, before you type anything. Every AI app has one running behind the scenes. It’s how the company tells the model to be helpful, polite, and so on.
Custom instructions are the version you control. Most AI assistants have a settings area where you can tell it about yourself and how you want it to respond: “I run a WordPress-based coaching business. Keep answers short. Use plain English. Never use corporate jargon.”
This is one of the easiest upgrades you can make. Five minutes of setup, and every conversation starts with the AI already knowing the basics about you.
Memory
Memory is the AI’s ability to remember things about you across separate conversations. You mention you have a membership site, and next week it still knows.
Different tools handle this differently. Some remember automatically, some only when you ask them to, and most let you view and delete what they’ve remembered. It’s convenient, but it’s also worth checking what’s in there every once in a while, both for privacy and because outdated memories can steer its answers in the wrong direction.
Memory is different from the context window. The context window is its short-term memory within one conversation. Memory is the long-term notebook it carries between conversations.
Projects
A project is a workspace for one ongoing job. You give it its own instructions and upload the relevant files, and every conversation inside that project has access to them.
For example: a “Newsletter” project with your voice guide, a few past issues, and instructions on your format. Every time you start a new chat in that project, the AI already has all of it. No re-explaining, no re-uploading.
ChatGPT, Claude and Gemini all have some version of this, though the names and details differ. It’s one of the most useful features for small business owners, and one of the least used.
Custom Assistants (Custom GPTs, Gems)
A custom assistant is a mini AI you set up for one specific purpose, with its own instructions and reference files, that you can reuse (and in some cases share with others).
OpenAI calls these custom GPTs. Google calls them Gems. The idea is the same: instead of explaining the task from scratch every time, you build a “Product Description Writer” or a “Support Reply Drafter” once, and just use it.
You’ll also see companies publish custom assistants for others to use, sort of like an app store. Some are useful. Many are just a set of instructions you could have written yourself.
Skills
A skill is a packaged set of instructions (and sometimes files or small scripts) that teaches an AI how to do one specific job your way. The AI loads a skill only when it’s needed for the task at hand.
Think of it like a standard operating procedure for your assistant. A “WordPress Blog Post” skill might include your formatting rules, your heading conventions, your call-to-action rules, and a checklist to run before publishing. When you ask the AI to prepare a post, it pulls in that skill and follows it.
Skills started with Anthropic’s Claude in late 2025, but they’ve since become an open standard that a growing list of tools support, including ChatGPT. That’s a big deal, because it means a skill you build isn’t necessarily locked into one company’s product.
The difference between a skill and a custom assistant is subtle but real. A custom assistant is a whole separate AI you switch to. A skill is more like a piece of know-how your main AI can reach for whenever it needs it. In my own business, I have dozens of these, and they’re a huge part of how AI actually does real work for me.
Knowledge Base and RAG
RAG stands for Retrieval-Augmented Generation. Terrible name. Simple idea.
Instead of relying only on what it learned in training, the AI first retrieves relevant information from a set of your documents, then uses it to generate its answer. It’s like letting it look at your files before it responds.
This is what powers things like “chat with your documents,” or a support chatbot on a website that answers questions using that company’s help articles.
You’ll often hear “vector database” in the same breath. That’s the storage behind it. Your documents get converted into long strings of numbers (called embeddings) that capture their meaning, so the system can find passages that are about the same thing as your question, even if they don’t use the same words. You don’t need to understand the mechanics. Just know that when a tool says it can “train on your content,” it’s usually doing RAG, not actually retraining the AI.
Fine-Tuning
Fine-tuning is actually retraining a model on your own examples so its behavior changes at a deeper level. It’s more technical, more expensive, and something you’d do through an API rather than a chat app.
For most solopreneurs, fine-tuning is overkill. Good custom instructions, projects, skills and RAG get you 95% of the way there, with none of the complexity. If someone’s pitching you fine-tuning for a small business use case, it’s worth asking why the simpler options won’t do the job.
Connecting It to Your Tools
This is where AI goes from “something you chat with” to “something that does work in your business.” It’s also where the vocabulary gets the most confusing, so let’s take it slow.
API
API stands for Application Programming Interface. It’s a way for one piece of software to talk to another directly, without a human clicking buttons.
Every major AI company offers an API. It’s how other apps build AI into their own products. That “AI writing assistant” button in some WordPress plugin? It’s almost certainly sending your text to OpenAI, Anthropic or Google through their API and showing you what comes back.
You don’t need to be a developer to care about this. APIs are why AI can show up inside the tools you already use, and they’re the plumbing behind most of the connections and automations in this section.
API Key
An API key is a long password-like string that identifies you when a piece of software talks to an AI company’s API. It’s also how you get billed.
If a WordPress plugin or other tool asks you to “paste your OpenAI API key,” this is what it wants. You’d create one in your account with the AI company, paste it in, and then pay for whatever that tool uses.
Treat API keys like passwords. Anyone who has your key can run up charges on your account. Don’t paste them into random websites, don’t email them around, and set a spending limit in your account if the provider allows it.
Pay-As-You-Go (API) vs. Subscription Plans
There are two completely different ways to pay for AI, and people mix them up constantly.
The subscription is the monthly plan for the chat app: ChatGPT Plus, Claude Pro, Google’s AI plans and so on. Flat fee, you use the app, done.
The API is pay-as-you-go, billed per token, and it’s separate. Your ChatGPT Plus subscription does not include API usage. So if a plugin needs an OpenAI API key, you’ll be paying for that separately, even if you already subscribe to ChatGPT.
For most everyday use, the subscription is simpler and a better deal. The API makes sense when another tool needs to use AI on your behalf.
Usage Limits (Rate Limits)
Every AI plan has limits on how much you can use it in a given period. Free plans have tight limits. Paid plans have higher ones. The most powerful models and features (like reasoning mode) tend to use up your allowance faster.
If you’ve ever seen “You’ve reached your limit, try again at 3pm,” that’s a usage limit. APIs have them too, usually called rate limits. It’s one of the main reasons people upgrade to a paid plan.
Integration and Connector
An integration (or connector) is a link between your AI assistant and another service… your email, your calendar, Google Drive, your project management tool, your WordPress site.
Once connected, the AI can read from that service, and depending on the permissions you give it, take actions there too. “What’s on my calendar Thursday?” “Find the email from my accountant last month.” “Draft a reply and put it in my drafts folder.”
This is the bridge to having AI do real work for you. It’s also where you need to be most careful. Start with read-only access where you can, and keep an eye on what you’ve connected. (I talked about the privacy side of this in the first post.)
MCP (Model Context Protocol)
MCP is the standard way AI assistants connect to other tools. You’ll see it everywhere now, so it’s worth understanding.
Here’s the easiest way to think about it: MCP is like a USB port for AI. Before USB, every device needed its own special plug. Before MCP, every AI company needed a custom-built connection for every tool. With MCP, a tool builds one connector (called an MCP server), and any AI assistant that speaks MCP can plug into it.
Anthropic created MCP in late 2024. It caught on fast. OpenAI and Google both adopted it, and it’s now maintained by an independent foundation rather than any one company.
Why it matters to you: when you see “MCP server available” on a tool you use, it means you can likely connect it to Claude, ChatGPT and others without any custom work. Plenty of WordPress tools are adding MCP support too, which means AI can work directly with your site.
Plugin
This one is confusing, because “plugin” means a few different things.
In WordPress, a plugin is software you install on your site to add features. Nothing to do with AI (unless it’s an AI plugin).
In the AI world, “plugin” has been used for different things over time. Early on, ChatGPT had “plugins” that let it reach outside services (those were later replaced). Today, in some tools, a plugin is a bundle… a package that installs a set of skills, connectors and settings all at once, so you can add a whole capability in one step.
The short version: in AI-land, when someone says plugin, ask “a plugin for what?” It’s an add-on of some kind, but the details depend completely on the tool.
Tool Use (Function Calling)
Tool use is the AI’s ability to decide on its own to use a tool mid-task: search the web, run a calculation, check your calendar, call an API, read a file.
You’ll also hear this called “function calling,” which is the more technical name. It’s the underlying skill that makes connectors, MCP and agents work. Without tool use, an AI can only talk. With it, it can act.
Agent (and “Agentic AI”)
An AI agent is an AI that works toward a goal on its own, taking multiple steps and using tools along the way, instead of just answering one question at a time.
The difference is easiest to see with an example. You ask a chatbot “how do I find broken links on my site?” and it explains how. You ask an agent “find the broken links on my site and give me a list,” and it actually goes and does it… crawls the pages, checks the links, and hands you the results.
“Agentic AI” is just the adjective form. You’ll see it used a lot in marketing, sometimes accurately and sometimes not. The real test: does it actually take actions and work through steps on its own, or is it just a chatbot with a fancier name?
This is where the real time savings are, and it’s where I think most solopreneurs need to end up. It’s also where caution matters most. An agent that can change things can also break things. Start small, keep backups, and approve its work before it goes anywhere.
Harness
A harness is the software an AI model runs inside of. It’s everything around the model that turns it from something that answers questions into something that can get work done: the loop that lets it take step after step, the tools it’s allowed to use (your files, a web browser, connectors), how it keeps track of what it’s doing, what it has to ask your permission for, and the behind-the-scenes instructions it follows.
The chat apps themselves are harnesses, just fairly simple ones. More powerful harnesses include tools like Claude Code, OpenAI’s Codex and Cursor, which let the AI work directly with files on your computer and carry out long, multi-step jobs.
Here’s why this matters: the same model can behave very differently depending on its harness. Claude in the regular chat window and Claude inside Claude Code are the same model. But in the chat window, it mostly talks. Inside Claude Code, it can read and edit files, run commands, and work through a 30-step task on its own. Same brain, very different workplace.
So when you’re comparing AI tools, don’t just ask which model they use. Ask what the harness lets it do. And the flip side: plenty of AI products you’ll be asked to pay for are a thin layer around one of the same models you already have access to. Before you add another AI subscription, ask what it does that your main AI assistant can’t.
Wondering which parts of your own business an agent could take off your plate? I built a free tool that sorts that out for you in about 10 minutes.
Computer Use and Browser Agents
A specific kind of agent that operates a computer or web browser the way you would. It looks at the screen, moves the mouse, clicks buttons, fills in forms.
This is useful for tools that don’t have an API or connector. The AI can just use the website like a person would. It’s powerful, but it’s slower and more error-prone than a proper connection, and you should be very careful what you let it log into. Never let it enter passwords or payment details on its own.
Automation (vs. Agent)
You may already use automation tools like Zapier or Make. Those follow fixed rules you set up: “When someone buys X, add them to list Y and send email Z.” Same steps, every time.
An agent is different. You give it a goal, and it figures out the steps. That makes agents more flexible, but also less predictable.
The two work well together. A lot of automation tools now let you drop an AI step into the middle of a workflow (“summarize this form submission and decide which category it belongs in”). For repetitive, predictable processes, plain automation is still often the better choice. It’s cheaper, faster and more reliable.
Where It Runs
Cloud AI vs. Local AI
Cloud AI runs on the AI company’s servers. When you use ChatGPT, Claude or Gemini, your prompt travels to their data center, gets processed there, and the answer comes back. This is how almost everyone uses AI, and it’s where the most powerful models are.
Local AI runs entirely on your own computer. Nothing leaves your machine, which makes it 100% private. The catch is that it takes serious hardware to run well, and the models you can run at home generally aren’t as capable as the big cloud ones. Free tools like Ollama and LM Studio make it much easier than it used to be.
Open Models (Open-Weight Models)
Models whose inner workings (the “weights,” meaning those learned parameters) are published for anyone to download and run. Meta’s Llama, and models from companies like Mistral, are well-known examples.
Open models are what make local AI possible. They also let other companies build products without depending on OpenAI or Google. You’ll sometimes see “open source” used loosely here, but “open-weight” is more accurate, since the training data and methods usually aren’t shared.
Buzzwords You’ll Hear Everywhere
A few more terms that get thrown around constantly. Some useful, some mostly hype.
Prompt Engineering
The skill of writing prompts that get good results. Early on, it was treated like a dark art, with people selling courses on secret “magic prompts.”
The truth is less mysterious. Models have gotten much better at understanding plain requests, so the tricks matter less. What still matters is the stuff that makes you good at delegating to a human: give context, be specific about what you want, show examples, and push back when the first answer isn’t right. That’s most of it.
Don’t worry about magic prompts. I usually just talk to my AI in plain English. Works just fine.
Vibe Coding
A term for building software by describing what you want to an AI and letting it write the code, without necessarily understanding the code yourself. You go on the “vibe,” check whether it works, and ask for changes.
It’s genuinely changed what non-programmers can build. You can now create a simple calculator, a custom WordPress snippet, or even a small web app by describing it. But there’s a big caveat for anything that touches your live site or customer data: code you don’t understand can have security holes and bugs you won’t spot. Test on a staging site, keep backups, and get a second pair of eyes on anything important.
Vibe coding is the future, but don’t make the mistake of thinking that it means you don’t need to know what you’re doing. AI can create crumby code. Seen it many times.
AI Slop
The internet’s name for low-effort, mass-produced AI content. Generic blog posts, bland social posts, weird AI images… stuff clearly churned out with no human thought behind it.
I bring it up because it’s the trap to avoid. AI can help you create more, faster. But if the result is slop, your readers will notice, and it’ll hurt you more than help. The fix is the same thing I keep coming back to in this series: AI for the grunt work, you for the thinking, the stories and the opinions.
AGI (Artificial General Intelligence)
The idea of an AI that can do basically any intellectual task a human can, at a human level or beyond. It’s the stated goal of several big AI companies, and a constant topic of hype and debate.
Here’s the thing: nobody agrees on exactly what AGI means, or how you’d know when it’s arrived. For running your business, it’s mostly a distraction. Focus on what today’s tools can actually do for you… which is already a lot.
Keep This Handy
That’s the vocabulary. You don’t need to memorize any of it. The point is that the next time someone drops “MCP” or “context window” or “agentic” into a conversation, you’ll know what they’re talking about… and whether it matters to you.
I’ll keep this post updated as new terms come along (and they will). If there’s a term you keep running into that isn’t here, let me know and I’ll add it.
Knowing the vocabulary is step one. Step two is figuring out where AI actually fits in your business. The AI Delegation Map walks you through the seven functions every business has and shows you what to hand off first.
Next up in this series, I’ll dig into the building blocks that all these AI tools share, like projects, skills and connectors, and what each company calls them. Because once you understand those, comparing the tools gets a whole lot easier.
