20 terms every AI literate person knows in 2026, defined like a human wrote them. Plus the 5 things to learn next once these click.
How to use this
Skim it once end-to-end. Then come back to it as a reference whenever a term shows up in a podcast, a launch, a meeting, or a Slack message and you want to actually understand what's being said.
Each term has a plain definition plus a real example showing how it shows up in actual work. That's the part that makes a term stick.
Models
The actual AI engine doing the work. Claude has 3 main models: Haiku (fastest, cheapest, good for simple tasks), Sonnet (the daily workhorse, handles 90% of work), and Opus (most powerful, used when accuracy matters more than speed). ChatGPT and Gemini have similar tiers with different names.
In the wild“Run this on Opus 4.7, the answer needs to be airtight.”
Tokens
Roughly one word of text. Both your input and the model's output count toward your usage. A 5,000-word doc is roughly 6,500 tokens. Bigger conversations use more tokens. When you hit a plan limit, this is what got used up.
In the wild“That doc is 50K tokens, it'll fit in one chat but it's going to slow Claude down.”
Context window
How much information the model can hold in working memory at one time. Bigger context windows mean you can paste longer documents and Claude remembers everything across a longer conversation. When you hit the limit, start a new chat or move to a Project (which has persistent memory).
In the wild“Claude has a 1M context window now, which means you can paste an entire book and ask questions about it.”
Hallucination
When AI generates a fabricated fact, source, or citation that looks real but isn't. Hallucinations happen most often on questions outside the model's training data, on legal or medical specifics, and on anything that needs a specific source citation. Always verify with a primary source.
In the wild“That study Claude cited doesn't exist, it hallucinated the URL. Always check.”
Knowledge cutoff
The date the model's training data was last updated. Anything that happened after the cutoff, the model doesn't natively know. To get current info, the model has to search the web in real time (which Claude can do via tools and connectors).
In the wild“Don't trust Claude for news from this week unless you let it browse the web. The training data has a cutoff.”
System prompt
The instructions Claude reads before every conversation. A regular prompt is your question. A system prompt is the standing orders. Custom GPTs, Claude Projects, and Cowork all use system prompts to set tone, role, and constraints.
In the wild“The system prompt tells Claude it's my career coach. Every chat in this Project starts there.”
Multi-modal
AI that can handle more than just text. Multi-modal models read images, listen to audio, watch video, and write in text. You can paste a screenshot, send an audio clip, or upload a video, and Claude can work with all of it.
In the wild“Send me a photo of the contract page, Claude can read it directly since it's multi-modal.”
Memory
When turned on, Claude remembers things across conversations. Your role, your preferences, recurring projects. It's the difference between an assistant who knows you and a chat window that forgets every time. Turn it on in Claude settings.
In the wild“I told Claude once that my husband's name is Tray. With Memory on, every relevant chat references him correctly.”
Connector
A 1-click bridge between Claude and an app you already use (Gmail, Notion, Canva, Stripe, etc.). Once installed, Claude can search inside that app, pull data from it, or take action in it. No code. 398+ official connectors are live right now.
In the wild“I installed the Notion connector. Now Claude can search every page and database in my workspace.”
Artifact
A live mini-app or document Claude builds in a panel right next to your chat. Spreadsheets, dashboards, interactive tools, simple websites, full prototypes. Editable, shareable via link, no developer needed.
In the wild“I asked Claude to build a hiring scorecard. The Artifact is a fully interactive form I can send to my team right now.”
Projects
A workspace inside Claude with persistent files, custom system prompt, and ongoing chat history. Think of it as a folder where you store everything related to one topic, role, or business. Claude remembers the context every time you open it.
In the wild“My ‘Daily Co-Pilot’ Project has my goals, my role, and 10 writing samples. Every chat is already on-brand.”
Skills
A saved task you trigger with a slash command. Instead of writing the same multi-step prompt over and over, you save it once and run it with /skillname. Pre-built Skills are in the directory. You can also build your own using Plugin Create.
In the wild“I built a /weekly skill. Every Friday I type /weekly, dump my wins, and get my status update in 30 seconds.”
Cowork plugin
A pre-built workflow stack that turns Claude into a specialist for one role. Install Sales, Claude does sales. Install Finance, Claude does bookkeeping. Each plugin comes with the skills, prompts, and connector preferences that role needs. 30+ official plugins available.
In the wild“I installed the Small Business plugin. It already knows how to chase invoices, run payroll, and triage leads.”
Extended Thinking
A mode where Claude reasons through a problem step-by-step before answering. The model essentially “shows its work” internally, which means better answers on math, logic, and nuanced reasoning tasks. Comes at the cost of being slower and using more tokens.
In the wild“Turn on Extended Thinking for this analysis. I need it to think it through, not just guess.”
Deep Research
A mode that runs a long, multi-step research session and returns a sourced, multi-page report. Claude reads multiple websites, cross-checks claims, and writes the synthesized output. Takes 10 to 30 minutes. Used for due diligence, market research, competitive analysis.
In the wild“Run Deep Research on the top 5 competitors in this category. Pull pricing, positioning, and recent funding.”
Chatbot vs Agent
A chatbot responds to a message. You ask, it answers. An agent takes action across your apps. You give it a goal, it plans the steps, executes them, and reports back. Chatbots talk. Agents do.
In the wild“The chatbot can draft the email. The agent can draft the email, send it, and follow up if no reply in 3 days.”
Agentic
The adjective for AI that can plan, take action, and complete multi-step tasks on its own. “Agentic AI” is the umbrella term for the wave of AI that doesn't just answer questions, it gets things done.
In the wild“The agentic future is already here. Most of my workflows are agentic now, not chat-based.”
Computer Use
A mode where Claude can see your screen and click, type, and navigate apps on your behalf. Used for tasks that don't have a clean API (filling forms, navigating Salesforce, scraping a dashboard, etc.). The closest thing to having a virtual assistant who actually uses your computer.
In the wild“I gave Claude Computer Use access to my browser. It updated 50 Notion pages overnight while I slept.”
Vibe coding
Building real software by describing what you want in plain English. Tools like Lovable, Replit Agent, Bolt, and v0 turn your description into working code, deployed and ready. No traditional coding required. The fastest path from idea to shipped app in 2026.
In the wild“I vibe-coded the entire customer portal in Lovable last weekend. Charged my client $3,500 for it.”
Fine-tuning vs Prompting
Fine-tuning actually modifies the model's weights to specialize it on a specific dataset (technical, expensive, mostly done by companies building products). Prompting just changes what you ask without touching the model. 99% of what you'll do is prompting. Fine-tuning is rare for most users.
In the wild“We didn't fine-tune anything. We just wrote a really good system prompt with 10 examples. Same result, way cheaper.”
Once you've locked in the 20 terms above, here's where to go deeper. Each one is an action, not just a concept.
Next 01
Build your own Skill
Stop using only pre-built Skills. Open Cowork → Customize → Plugin Create. Describe a task you do every week (meeting recap, weekly status, client follow-up). The wizard walks you through it. By the end you have your own /command that runs the task in one click.
Next 02
Build your own Cowork agent
An agent is bigger than a Skill. It's a full role you've trained Claude to be (your inbox triage agent, your content repurpose agent, your sales-research agent). Build one for the most repetitive part of your week. From now on, that part runs itself.
Next 03
Ship a live Artifact
Ask Claude to build something interactive (a calculator, a tracker, a quiz, a one-pager). Share the Artifact link in real life. Watch what happens. Most people don't realize you can hand someone a working tool from a single Claude chat.
Next 04
Turn on Memory and curate it
Go to Claude settings. Turn on Memory. Then in your next 5 chats, intentionally feed it your role, your goals, your preferences, your style. After a week, Claude is operating with full context every time you open it.
Next 05
Set up Computer Use for one workflow
Pick one repetitive task that involves clicking around a browser (updating CRM records, transferring data between tools, refreshing dashboards). Set up Computer Use to handle it. Watch Claude do your work while you're in a meeting. It's the closest thing to magic AI offers right now.
If you haven't done the 7-day plan yet
If most of these terms are still new to you, start with Become AI Fluent in 7 Days. It walks through the foundations in 30 to 60 minutes a day, with real actions on each one. By Sunday these terms aren't theory, they're tools you've used.
For Your Job
If you’re ready to set up Claude for your specific job, with custom skills, connectors, and automations built around the work you do every day, I built a bootcamp just for you.
Start the Weekend Bootcamp →Get The Agents
These are the exact AI agents I have running across my businesses, automating 90% of my day-to-day operations. Install the same ones I use, ready to plug straight into your work.
Get the AI Agent Bundle →Work With Me
If you want help implementing AI into your business, I’m now taking on consulting work and would love to help you. Shoot me an email with your budget, scope, and a little about your business.
Email Me →© 2026 Mariah Brunner. All rights reserved.