Skip to main content

AI Glossary: Terms From the Workshop

This AI Glossary is Provided by DTC Inc. | Companion to the Working With AI workshop | Version 1.0 | September 2026

Plain-English definitions of the terms used during the AI workshop. These are the words that came up in the room, defined the way they were explained there. No prior technical knowledge assumed.

If a term you heard is missing, ask and it will be added.

🧠 How AI Works

Term What it means
AI Software that recognizes patterns in an enormous amount of human writing and uses those patterns to respond to you. It was not programmed with a list of answers. Intelligence emerged from the patterns.
Model The actual AI brain underneath whatever app you are using. Claude, ChatGPT and Gemini each run on their own models, and each company releases new versions regularly. Switching models inside an app is normal and does not lose your conversation.
LLM (large language model) The technical name for this kind of AI. "Language model" because it works with words. "Large" because of the volume of text it learned from.
Token Roughly, a word. AI works by weighing which words relate to which other words, then choosing what comes next.
Probabilistic The answer is chosen by weighted likelihood, not by a fixed rule. This is why AI is not a calculator.
Deterministic The opposite: the same input always produces exactly the same output. Spreadsheets and traditional software are deterministic. AI is not.
Noise A small amount of deliberate randomness, roughly 5 to 20 percent depending on the model, added so responses vary. It is why the same prompt twice gives you two differently worded answers. Two plus two still equals four; the sentence around the four changes.
Confidently wrong AI sounds exactly the same whether it is right or wrong. There is no tone change to warn you. This is the single most important thing to remember.

💬 Talking to AI

Term What it means
Prompt Whatever you type or say to the AI. That is all it means.
Context Everything the AI knows going into your request: who you are, what is happening, who is involved, what already happened, and what constraints exist. More context, better result. This was the central idea of the workshop.
Master prompt A standing block of information about you that gets injected into the conversation before it starts, every time. Picture a short book about you being handed to the AI before you say a word: your role, your preferences, your history, how you want to be spoken to. It is why one person's AI already sounds like them and a brand new account sounds generic. You do not write it from scratch. You ask the AI to interview you and build it.
System prompt The same idea, set at the organization or app level rather than by you personally. It is how a group can make everyone's AI follow the same file naming or structure rules.
Custom instructions Where most apps let you store your master prompt. Look in settings for "custom instructions," "personalization," or similar.
Dump Workshop shorthand for telling the AI everything, raw and unedited, without cleaning it up first. Bad grammar, half sentences and all. Dumping works better than writing something polished.
Intent What you are actually trying to accomplish, as opposed to the task you are asking for. "Write an email" is a task. "I need this person to understand why we changed the process and reply by Friday" is intent.
Interview me A phrase worth memorizing. Instead of asking a question directly, say: "Act as an expert in this and ask me every question you need to accomplish it." One person in the room got 86 questions back, and a far better result than they would have gotten otherwise.
Voice transcription Talking to the AI instead of typing. Most apps have a microphone button. Several people in the room found this the single biggest change to how much they use it.

📓 Memory and History

Term What it means
Memory The AI's stored knowledge of you, built up from your past conversations. It is what makes the AI feel like it knows you. It does not transfer between different AI tools.
Journal A running, time-ordered record of what you did, decided, and learned. You talk through your day, the AI writes it down, and from then on it answers from your actual history instead of guessing. Ten minutes a day.
Chronicle The same thing as a journal. Some people in the room use this word instead. Call it a journal, a chronicle, a log, or a sketch pad. What matters is that it is time-based and you keep adding to it.
Identity A named AI set up with its own memory, journal and knowledge base, so it stays consistent across conversations rather than starting fresh each time. Nate works with one called Pia; DTC has one called Apis. Underneath, both are the same model.
Knowledge base A structured library of an organization's documented processes and reference material that the AI can read from. DTC's runs on a system called BookStack, which is where this glossary lives.

🔌 Tools, Connections and Automation

Term What it means
Harness The app wrapped around the AI: what it can see, what it can touch, and how you talk to it. The Gemini panel in Gmail is a harness. The ChatGPT app on your phone is a harness. Same underlying AI, different harness.
Connector A link that gives your AI access to one of your other systems, such as your Drive, your email, or your calendar. It will ask your permission before connecting.
Agent An AI that does work on its own rather than just answering you. "Sort my inbox by rules I define" is an agent. In practice, an AI plus a connector to a system it can act in.
Agentic The adjective. An agentic setup keeps working without you watching. It will keep going until you stop it, which is exactly why guardrails matter.
Collision What happens when two AI agents, or the same AI in two places, work from different versions of the truth. You get contradictions and output that reads like garbage. The fix is a single live source both can read, rather than telling each one separately. As Nate put it: it is a reflection of how well organized you are.
Guardrails The limits you deliberately put on what an AI can see and do. Connecting more makes it more useful and also raises the stakes, so you choose the boundaries on purpose.
Automation A process that runs the same way every single time. AI is not automation, because it varies by design. The right move is to have AI build the automation, rather than asking AI to redo the same chore by hand every day.

🧰 Tools Named in the Room

Name What it is
Gemini Google's AI, built into Google Workspace. Its advantage is that it is already connected to your Gmail, Drive, Docs and Calendar with no setup. Open it from the apps menu; it looks much like ChatGPT, with your chats down the left.
ChatGPT OpenAI's AI. A broad working environment with memory, projects, files, voice and images.
Claude Anthropic's AI, and the model DTC uses. Strong with long documents, writing, analysis and building things.
Copilot Microsoft's AI, built into Microsoft 365 the way Gemini is built into Google Workspace.
NotebookLM A Google research tool. You give it your own sources and it answers only from them, with citations, then turns them into a podcast-style audio summary you can listen to. Several people left planning to try this first.
Cowork A mode where the AI works in your files and apps on your computer while you do something else. No coding involved.
Claude Code A tool that builds things: scripts, tools and automations. This is the deep end, mentioned so you know it exists.
CLI (command line interface) Typing commands to a computer rather than clicking buttons. How Claude Code is used. Nobody needs this to get value from AI.
Codex OpenAI's equivalent of Claude Code.

🗣️ Other Words You Heard

Term What it means
Deck A slide presentation. Not an AI term at all, and a fair thing to ask about.
Prompt reuse Pasting the same prompt again to get the same result. It will not give you the same result. If you need identical output every time, you want automation, not AI.
AI overwhelm The feeling of drowning in tools, each promising to be the one. The way out is to pick one and go deep rather than chasing every new release.
Free vs paid Paid accounts generally get better models, more memory and more features. Whether that is worth it depends entirely on how much you actually use it.

⚖️ What Not to Give It

This came up repeatedly and belongs in any glossary of this material.

  • Never share credentials. No passwords, no bank logins, no account access details, no Social Security number, no credit card numbers.
  • Be careful with other people's information. Health and care details, the people you serve and their families, employee and caregiver matters, immigration or legal status, and anyone's private financial affairs.
  • Know which account you are in. Work information belongs in the work account. When in doubt, strip out names and identifying details; the AI usually does not need them to help.
  • You stay accountable. AI can inform, suggest, draft, question, organize and analyze. Judgment, decisions and responsibility remain yours. If your name is on it, you have read every line of it.

💡 The one-line version: AI is a reflection of what you give it. Context is the whole game.

📞 Questions about any term here? Contact DTC at support@dtctoday.com. See also the companion resources: Working With AI: A Practical Guide and A Reflection of Us: AI Workshop Slides.

Back to top