The Frameworks
The 4Ds
Four skills for working with AI: deciding what to hand off (Delegation), communicating clearly (Description), evaluating what comes back (Discernment), and using AI responsibly (Diligence).
Why it matters: These are the core competencies the pilot develops in you.
Delegation
Deciding which tasks to give Claude and which to keep for yourself. This is where you start every interaction: does this task benefit from AI assistance, and what part specifically?
Why it matters: Smart delegation multiplies your time without sacrificing quality or control.
Description
How clearly you communicate what you need from Claude, including what you want, how you want it done, and what tone or format to use.
Why it matters: Better descriptions produce better outputs on the first try, which saves you revision time.
Discernment
Your ability to evaluate whether Claude's output is accurate, appropriate, and ready to use. This is the skill that prevents errors from reaching students, funders, or the board.
Why it matters: You're the gatekeeper. Your judgment is what makes AI safe in your organization.
Diligence
Using AI responsibly: being transparent about AI's role in your work, following governance rules, and owning every output you send.
Why it matters: Your name goes on it, so you verify it. Diligence protects your credibility and the organization's reputation.
Automation
When you give Claude a specific task with clear instructions and it executes. Think: 'Reformat this list into a table' or 'Summarize these meeting notes.' You define the task; Claude does it.
Why it matters: Automation handles the routine, freeing you for higher-impact work.
Augmentation
When you and Claude think together, going back and forth to develop something. Think: brainstorming grant angles, refining a draft through multiple rounds, or exploring different framings for a board presentation. You both contribute.
Why it matters: Augmentation is the back-and-forth where the real work happens: you bring the judgment and the context, and Claude helps you get further, faster.
Agency
When you set Claude up with background knowledge and guidelines so it can work more independently. Think: setting up Projects, Skills, and standing instructions so your context is already there instead of retyped each time. Claude does not remember across chats unless memory is turned on — Agency is about what you configure, not what Claude recalls.
Why it matters: Configuration you do once pays off in every conversation after it.
Agent / Agentic
Claude working in steps — reading, deciding, acting — with your supervision. Week 6 covers this mode of work.
Why it matters: Multi-step work is where supervision matters most. You stay the reviewer at every step.
Technical Terms You'll Hear
Prompt
The message or instruction you type to Claude. Everything you write in the chat box is your prompt, including questions, instructions, and any documents you paste in.
Why it matters: Your prompt is how you talk to Claude. The better your prompt, the better your result.
Prompt Engineering
The practice of writing prompts that get you better results. DC CAP uses the five-part brief to structure prompts: Role, Context, Task, Example, Format (Weeks 0 and 1 teach it).
Why it matters: Good prompt engineering means fewer revision rounds and faster results.
Iteration
Revising your prompt or pushing back on Claude's output to get a better result. The people who get the most from Claude are the ones who push back instead of accepting the first draft.
Why it matters: The more you iterate, the faster your own skill grows. Claude does not learn between chats; each revision sharpens what this conversation gives you.
Context Window
The amount of text Claude can hold in its memory during a single conversation. Think of it as Claude's working memory. If a conversation gets very long, Claude may lose track of things mentioned early on.
Why it matters: Managing it is the one skill the whole course builds: knowing what to load, when to start fresh, and what to carry forward.
Hallucination
When Claude states something confidently that sounds plausible but is factually wrong. This happens because Claude generates language patterns, not truth.
Why it matters: Always verify statistics, citations, and specific claims before using them. This is where Discernment saves the day.
Token
A small chunk of text, usually a piece of a word, that Claude processes (course rule of thumb: picture "next word"). You don't need to manage tokens yourself.
Why it matters: Just know that longer conversations and documents use more of Claude's working memory. It helps explain why very long chats can slow down.
Model
The AI system itself. Claude is a model built by Anthropic. Different models have different capabilities. DC CAP uses Claude, specifically the enterprise version with data protections.
Why it matters: When someone says 'model,' they mean the AI engine you're talking to. It's helpful to know what model you're using and its strengths.
LLM / Large Language Model
A model that predicts the next word from patterns learned in a huge amount of text. Claude is one.
Why it matters: Knowing it predicts patterns, not truth, explains both its strengths and why you verify.
System Prompt / Project Instructions
Background instructions that shape how Claude responds to you, set before the conversation starts. DC CAP's organizational context, brand voice rules, and governance policies live in Project Instructions so every team member gets consistent, context-rich outputs.
Why it matters: These are the rules and guardrails that make Claude behave like "DC CAP's Claude," not generic Claude.
Project (in Claude)
A workspace in Claude where you can load documents and instructions that persist across conversations. Your team can keep a shared Project holding reference documents and standing instructions so Claude has organizational context whenever you need it.
Why it matters: Projects eliminate the need to re-explain context every time. Claude remembers your organization's priorities and materials.
Skill (in Claude)
A specialized instruction set that tells Claude how to do a specific task well. DC CAP has built Skills for grant writing, student outreach, data interpretation, and more. Skills encode your team's best practices so Claude follows them every time.
Why it matters: Skills make your team's expertise repeatable and consistent. Once you build a Skill, anyone can use it—no training needed.
Memory (in Claude)
Short, stable facts Claude may carry between chats where the feature is enabled. You curate the list yourself in Settings.
Why it matters: Memory is opt-in and editable — you decide what Claude keeps, and you can remove anything.
Cowork
The Desktop App mode where Claude makes real files in a folder you point it at. Not just chat — actual documents, spreadsheets, and pages land on your machine.
Why it matters: Cowork turns a conversation into finished files you can open, edit, and send.
Connector
A deliberately opened, limited door that lets Claude read from a tool you already use. You choose what opens and how far.
Why it matters: Connectors give Claude real context from your tools without handing over everything.
Data Tier
DC CAP's four-level classification of what may enter Claude. Tier 1 — anything identifying a scholar or family — never enters.
Why it matters: The tiers are the governance rule you check before pasting anything in. See the Governance page for the full framework.
Capstone
The optional end-of-course build: a Skill or Project a teammate can run. It closes the pilot by turning what you learned into something your team keeps.
Why it matters: The capstone is where personal fluency becomes a tool the whole team inherits.