Here's the answer: about twenty terms cover everything you'll actually encounter, and each one fits in a sentence. A model is the trained AI itself, and bigger isn't always better. A context window is the working memory of one conversation, which is why long chats drift and fresh chats fix it. A hallucination is a confident-sounding wrong answer, which is why verify-what-matters is not optional. A prompt is any message you send. A brief is a prompt with direction, and the difference is your whole edge.
Then the platform layer. A Project is a workspace where files, instructions, and chats share context permanently, the end of re-explaining yourself. Memory is what carries across conversations, and it's yours to review and edit. A connector links Claude to your real tools like email and calendar. A skill is your way of doing something, made repeatable. A scheduled task is work that fires on a calendar without you present. And the distinction that matters most: automation runs your procedure, autonomy makes choices, and you grant the second slowly.
You'll know the vocabulary has done its job when sentences like these are your natural speech: start a fresh chat, this one's drifting. Put that in the Project so we stop re-explaining. That needs a brief, not a prompt. Verify that stat.
And carry one rule everywhere: anyone using jargon at you without a plain-language version either doesn't understand it or wants you not to. Insist on the one-sentence definition, especially from people selling things.
The full glossary is Volume 4 of our Claude Fluency series.
The jargon wall comes down in an evening.