Sub Lesson 5
A glossary for working with AI
Look up the terms used throughout the course
Brief explanations of the terms used in the course. You do not need to read them in order. Look up a term when you need it.
Asking AI directly is often quicker. A question such as “What does context mean?” can provide an explanation while you work.
Terms for organizing work
Chapters 1 and 2 introduce the concepts used throughout the course.
| Term | Meaning |
|---|---|
| Deliverable | Something produced: a document, report, post or workflow |
| Result | A change toward a desired state. A proposal is a deliverable; a customer deciding to proceed after reading it is a result |
| Desired state | What you want to be true when the Goal is achieved |
| Current state | What is actually true now, described with evidence |
| Difference | What remains between the current and desired states. Work resolves that difference |
| Main constraint | The most important obstacle to progress; choose one and act on it |
| ToDo | One action that returns a result you can assess |
| Goal | A unit of work that keeps the desired state and current state together |
| Child Goal | A smaller Goal within a larger Goal |
| Leaf Goal | A Goal without children, where the next action is performed |
| Goal Tree | The parent-child structure of Goals |
| Goal Loop | The Goal achievement loop and Goal update loop together |
| Goal achievement loop | Update the current state from results, then choose the next action |
| Goal update loop | Learn from results and revise the desired state itself |
AI tools
| Term | Meaning |
|---|---|
| Chat AI | AI such as ChatGPT that answers a request while you perform the resulting actions |
| AI agent | AI that can carry out actions as well as answer, such as Claude Code |
| Claude Code | Anthropic's agent for reading and writing files and running commands on your computer. See Supplementary lesson 1 |
| Codex / Cursor | Other tools serving a similar role. The course's approach applies whichever you use |
| Model | The AI system doing the reasoning, such as Fable 5. More capable models can handle larger delegated Goals |
| Prompt | An instruction to AI. Providing the Goal and current state often affects results more than refining the wording |
| Context | Background available to AI: documents, history, the current work state and knowledge you record |
| Token | A unit used to represent text for AI. A Japanese character is roughly around one token |
| Context window | The limit on how much AI can read at once. Conversation context is supplied again for each reply. See Supplementary lesson 4 |
| Usage-based / fixed monthly pricing | Paying for consumption or a fixed subscription. A fixed monthly cost can make it easier to delegate a wider range of work |
| Terminal | A text interface for entering commands. The desktop app lets you begin without learning terminal operations |
Connections
An AI working with local files begins with file access. External connections add capabilities, as explained in Chapter 4.
| Term | Meaning |
|---|---|
| MCP | A shared protocol for connecting AI to external tools. Supporting tools can use the same connection approach |
| Connector | An available integration in an app, such as Gmail or a calendar, that you select and authorize |
| API | An interface a service provides for accessing data or operations, often authenticated with an API key |
| CLI | A way to operate a tool through commands; some work can take fewer steps than through MCP |
The methods have different names, but each adds a capability AI can use. Learn how to connect and disconnect the tools needed for your work.
Execution
| Term | Meaning |
|---|---|
| Automation | Running a fixed procedure at a specified time or event, with consistent behavior |
| Autonomous work | AI uses the Goal and current state to choose the next action when no fixed procedure covers the work |
| Routine | A mechanism for recurring actions at specified times; AI can help configure it |
| Scheduled execution | Starting AI at a time such as 9 a.m. daily, Monday each week or month-end |
| Event trigger | Starting AI when something happens, such as an email, approval or new file |
| Local execution | Running on your own computer, which must be on for the process to run |
| Cloud execution | Running independently of your computer; move to it after the process works without corrections |
| Stopping conditions | Conditions defined before autonomous execution: missing information, scope limits and unexpected situations |
| Harness | The instructions, tools and procedures that support an AI agent's work. It is a means of working, not a result |
Working environment
| Term | Meaning |
|---|---|
| Workspace | The folder where AI works; opening a folder establishes the working location |
| CLAUDE.md / AGENTS.md | An instruction file read when AI begins in a folder. Keep it to source locations and working rules |
| Skill | A reusable procedure stored where AI can read it when relevant |
AI behavior
| Term | Meaning |
|---|---|
| Sycophancy | Constructing reasons to agree with a user's stated view, even when it is wrong |
| Hallucination | Presenting invented or unsupported information in a confident response |
| Reward hacking | Satisfying the stated objective by changing or exploiting the measurement, such as rewriting a test to make it pass |
Read Supplementary lesson 4: Understand how AI behaves for examples and practical responses.
For terms used in Addness screens and features, including Today's ToDos, archiving, streaks and permissions, use the Help Center glossary.