Every AI term.
In plain English.

Agent harness The harness is the app wrapped around a model that gives it tools, memory, and the ability to act. The model… all levels Agent loop The agent loop is how an AI agent works through a task. It runs in a cycle: observe, think, act. It observes… all levels AI agent An AI agent is an AI you hand real work to. The cleanest way to think about it: a Chat model is question to… all levels AI slop AI slop is the generic, low-quality, obviously-AI output you get when nothing real is guiding the model.… all levels Chat model A chat model is the ask-a-question-get-an-answer way of using AI. ChatGPT, Claude chat, that's what most… beginner Context window The context window is how much your AI can hold in its head in one chat before it loses the plot. Picture a… all levels Hallucination Hallucination is when a model makes things up or starts forgetting, confidently telling you something that… all levels Heartbeat A heartbeat is a scheduling loop that keeps an always-on agent ticking. The agent wakes itself on an… builder LLM The LLM is the brain. It stands for Large Language Model, and it's just the Chat model, Claude or ChatGPT or… all levels Local model A local model is an AI you run on your own machine instead of someone else's. Claude, ChatGPT, and Gemini all… all levels Markdown Markdown is a plain-text file format that AI agent read natively. An MD file is just a markdown file, a… all levels Model-agnostic Model-agnostic means building your setup so it isn't tied to one model or one company. You can run Claude for… builder Multimodal Multimodal just means an LLM with eyes. Before, AI mostly worked on text. Now it can actually see. all levels Open-source model An open-source model is one where the company releases the weights (the actual trained brain) for anyone to… builder Prompt injection Prompt injection is when malicious instructions get smuggled into your AI agent through something you… builder Second brain A second brain is all of your AI agent's context kept as local Markdown files in one central place. Every… builder Tool calls A tool call is the AI agent actually doing something instead of just talking. Searching the web, running… all levels Trigger A trigger is whatever kicks off an automated run. You set one up when you build a routine, and from then on… builder

Agent teams Agent teams are multiple fully independent Claude instances coordinating directly with each other. They share… builder Computer use Computer use is when the model gets its own computer, or takes control of yours. As Remy puts it, the model… all levels Context compaction Context compaction is Claude tidying up the conversation as it grows, so a long session doesn't hit a wall… all levels Dynamic Workflows Dynamic Workflows is an Anthropic feature that takes one big job, splits it across a bunch of Subagents that… builder Persistent memory This is what people actually mean when they talk about AI with memory. There's no magic feature built into… all levels Plan mode Plan mode stops the AI agent from touching any files so it can only talk through the work first. No writing,… all levels Plugin A plugin is an add-on you install to give your agent new abilities. Where a single Skills teaches it one… all levels Reasoning effort level Reasoning effort level is a dial for how hard a model thinks before it answers. Some models are… builder Remote control Remote control is running and watching your Claude Code sessions from your phone. Claude keeps working on… builder Security scanner The security scanner is a built-in Claude Code command that reads through your codebase looking for… builder Skills Skills are reusable instruction manuals that teach your AI agent how to do a task. They're SOPs for your AI.… all levels Subagents Subagents are secondary AI workers your main AI agent spins up to handle the grunt work: reading files,… builder

Context engineering Instead of writing one big clever prompt every time, you build up a set of reusable files about your… all levels JSON prompting JSON prompting means writing your prompt as structured data instead of a paragraph. Rather than describing… builder Multi-agent orchestration Multi-agent orchestration is running a bunch of AI agent together under one view, with shared context,… builder Prompt engineering Prompt engineering is crafting the wording of a single request to get a better answer out of an AI. How you… all levels Vibe coding Vibe coding is building software or websites by describing what you want to an AI in plain English and… all levels Voice cloning Voice cloning is recreating your own voice in AI, so you can generate narration and voiceovers that sound… all levels Web scraping Web scraping is pulling the actual content off a web page. Claude can search the web on its own, but it can't… builder

AI employees The simplest way to think about an AI employee: it's an AI persona set up to do one role in your business,… all levels Broad agents, niche skills This is the architecture rule I keep coming back to: build a few broad, capable agents that each own a whole… builder Orchestrator skill An orchestrator skill is a Skills that calls several other skills in sequence to run a whole multi-step… builder Scheduled tasks Scheduled tasks are recurring agent runs that fire on a timer instead of you kicking them off. Hourly, daily,… all levels Vibe design Vibe design does for branding and design what Vibe coding does for code. You describe what you want and the… all levels Voice-to-text Dictation apps that let you talk to your AI agent instead of typing. You speak, it types. That's the whole… all levels

Claude Code Anthropic's CLI / Terminal-based agent that runs locally on your machine. It's my daily workhorse, and I run… all levels Claude Cowork Claude Code wrapped in a clean interface for non-coding work. It's aimed at the non-technical founder who… all levels CLI / Terminal The terminal is a text-based way to control your computer by typing commands instead of clicking around. It's… all levels GitHub GitHub is just Google Drive for code. It's where people store and back up their code online, the same way… beginner n8n / Make Workflow automation tools. You wire up a rigid backend process once and it runs itself: someone buys a… all levels NotebookLM Google's AI learning tool. Each notebook is a subject you want to learn about. You upload your source… all levels Obsidian A free note-taking app where every note is a plain Markdown file sitting on your computer. Nothing is locked… all levels

API A programmatic doorway that lets you plug a model or service into an app, or automate a workflow. It's the… builder CLAUDE.md / AGENTS.md The instruction file your agent reads at the start of every session before it does anything. It tells the… builder Composio Composio is one Connectors that holds all your other tool connections in one place. You connect every tool… builder Connectors Connectors are pre-built integrations that plug an agent into your outside apps: Gmail, Slack, Drive, Notion,… all levels Cron A time-based job scheduler. It's the timer behind anything that runs on its own clock: every hour, every… builder Self-hosting Self-hosting means running your whole setup, or a single agent, on your own hardware or your own rented… builder VPS A VPS is a Virtual Private Server: a remote, always-on computer you rent and leave running in a data centre… builder Webhook An automatic Trigger that fires the moment something happens. A message arrives, a payment lands, and the… builder

Embedding model An embedding model turns your content into something you can search by meaning instead of exact words. Text,… builder Model tiers (Haiku, Sonnet, Opus) Claude comes in three tiers, and the names tell you the tradeoff. Haiku is the cheapest and fastest. Sonnet… all levels

MCP (Model Context Protocol) MCP is how an AI agent talks to and acts inside your other apps and data, like Stripe, Gmail, or Notion.… all levels OAuth OAuth is the "log in with Google" pattern, but for your agent's tools. Instead of hunting down an API and… builder