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.
Skim it once, then come back whenever a term pops up in a podcast or meeting. Each one has a plain definition and a real example.
01
Models
The AI engine doing the work. Claude has four. Haiku is the fastest and Sonnet is fast and smart. Opus is the everyday heavy hitter, and Fable is the most powerful.
In the wild: “Run this on Fable, the answer needs to be airtight.”
02
Tokens
The chunks AI reads and writes in. On current Claude models a token is about half a word, so a 5,000-word doc is roughly 9,000 tokens. Input and reply both count toward usage.
In the wild: “That doc is 50K tokens. It fits, but it’ll eat my usage.”
03
Context window
How much the model holds in working memory at once. On paid plans, current Claude models take about 1 million tokens, hundreds of thousands of words.
In the wild: “A 1M context window means I can drop in the whole contract and ask questions.”
04
Hallucination
When AI makes up a fact or citation that looks real. It happens most on niche details and exact sources, so check anything important.
In the wild: “That study it cited doesn’t exist. Always click the link.”
05
Training cutoff
The date the model’s training data stops. For anything newer, it has to search the web.
In the wild: “Turn on web search for this week’s news. Its training has a cutoff.”
06
System prompt
Standing orders the AI reads before your message. Your prompt is the question; the system prompt sets the role and rules. Claude Project instructions work this way.
In the wild: “The Project instructions tell Claude it’s my career coach, so every chat starts there.”
07
Multimodal
AI that handles more than text. Claude reads images, screenshots and PDFs. Gemini can also watch video.
In the wild: “Just send a photo of the contract page. Claude can read it.”
08
Memory
Claude remembers useful things across chats, like your role and projects. It’s on by default on personal plans. Find it in Settings > Your Claude.
In the wild: “I told Claude once that my husband’s name is Tray. Now it just knows.”
09
Connector
A bridge between Claude and an app you use, like Gmail or Notion. Claude can then search it or take action in it, and it asks before acting.
In the wild: “I connected Notion. Now Claude can search every page in my workspace.”
10
Artifact
A document or working mini-app Claude builds next to your chat, like a tracker or calculator. You can share it with a link.
In the wild: “I asked for a hiring scorecard and got a working form I can send my team.”
11
Projects
A workspace with its own files, instructions and chats. Every chat inside starts with that context. Free accounts get up to five.
In the wild: “My ‘Daily Co-Pilot’ Project has my goals and 10 writing samples. Every chat is on-brand.”
12
Skills
Saved instructions for a task you repeat. Build it once and Claude loads it on its own whenever a request matches.
In the wild: “I built a weekly-update skill. I dump my wins and get my status report in 30 seconds.”
13
Plugin
A bundle of skills and connectors that makes Claude a specialist for one role. Install them from the Plugins page (formerly Customize).
In the wild: “I installed the Small Business plugin. It already knows how to chase invoices and triage leads.”
14
Thinking and effort
Newer models reason before they answer. On Claude’s latest models thinking is always on, and effort is the dial. More effort means deeper but slower answers.
In the wild: “Crank the effort up for this one. I need it reasoned through, not guessed.”
15
Research
Claude’s deep research mode. It runs many searches and returns a cited report, usually in 5 to 15 minutes. On paid plans, click +, then Research.
In the wild: “Run Research on the top 5 competitors. Pricing, positioning, recent funding.”
16
Chatbot vs agent
A chatbot answers. An agent takes a goal, plans the steps, does them in your apps and reports back.
In the wild: “The chatbot drafts the email. The agent sends it and follows up in 3 days.”
17
Agentic
Describes AI that plans and finishes multi-step tasks by itself.
In the wild: “Most of my workflows are agentic now. I set the goal and check the result.”
18
Computer use
AI that sees a screen and clicks and types for you. In Claude that’s Claude in Chrome on paid plans, plus computer use in the desktop app (a research preview on Pro and Max).
In the wild: “Claude in Chrome updated 50 CRM records while I was in a meeting.”
19
Vibe coding
Building software by describing it in plain English. Lovable, Replit, Bolt, v0 and Claude Code turn the description into working code.
In the wild: “I vibe-coded a client portal in Lovable last weekend.”
20
Fine-tuning vs prompting
Fine-tuning retrains a model on your data, which is costly and mostly done by companies. Prompting just changes what you ask. You’ll mostly prompt.
In the wild: “We skipped fine-tuning. A great system prompt with 10 examples did the job.”
01
Build your own skill
Open Customize > Skills (on the new Plugins page, filter to skills), click +, then Create skill. Or just tell Claude “help me turn my weekly status update into a skill.”
02
Build your first agent
An agent is a Project with clear instructions plus the connectors and skills it needs. Start with your most repetitive task. My first agent guide walks you through it.
03
Ship a live Artifact
Ask Claude to build something interactive, like a tracker or quiz, and share the link with a real person.
04
Curate your memory
In Settings > Your Claude, check that Memory from chats is on, then fill in Instructions for Claude with your role, goals and style. After a week, review what it remembers and delete anything wrong.
05
Try computer use on one task
Pick one repetitive browser task, like updating CRM records. Try it with Claude in Chrome and watch the first runs before trusting it alone.
If most of these terms are new, start with Become AI Fluent in 7 Days. By day 7 they’re tools you’ve used.