Your Quick Reference for Claude at DC CAP
Enterprise AI Leadership Pilot · April 2026
Read the six sections below and the AI Governance Framework. Together they cover how Claude works at DC CAP — interfaces, files, skills, data handling, approved use cases, and responsible use.
Complete the AI Fluency Assessment. Takes about 8 minutes. Captures your baseline so we can measure growth. Includes a governance acknowledgment.
Finish two free Anthropic Academy courses: Claude 101 (basics of working with Claude) and AI Fluency for Nonprofits (4D framework for mission-driven work). Both earn certificates.
claude.ai
Your everyday workspace
Key feature: Conversations are private by default. Use Projects (Section 2) for shared work.
"Think of Chat as your personal thinking partner."
Desktop App
Your power tool for deliverables
Key feature: Claude creates actual files, not just text. Skills encode DC CAP's organizational knowledge.
"Think of Cowork as your production studio."
CLI
For technical teams
Claude Enterprise gives DC CAP three collaboration layers. Each one serves a specific purpose, and understanding these layers is the foundation for working effectively as a team.
This is your collaboration center.
A Project is a persistent shared workspace. Every conversation inside a Project starts with the same organizational context — DC CAP's policies, your team's guidelines, reference documents, past work. Claude remembers all of it across every conversation, so you never re-explain who DC CAP is or what your program does.
Your teammates can see conversations in the same Project, continue work someone else started, and build on each other's outputs. When you upload a document to a Project's knowledge base, every team member's Claude conversations benefit from it immediately.
What goes here: Program documentation, grant templates, evaluation frameworks, policy references, student success protocols, strategic plans — anything your team references repeatedly.
This is your production workspace.
Cowork connects Claude to a folder on your computer and gives it the ability to create real files: Word documents, PowerPoints, spreadsheets, PDFs. It also connects to external tools through connectors (Monday.com, Google Drive, Canva). When you need a polished deliverable or need to work across multiple tools, Cowork is the right interface.
Cowork also runs DC CAP's organizational Skills — pre-built expert workflows that encode our voice, framing rules, and policy accuracy. You describe what you need in plain language; Cowork detects the right Skill automatically.
What happens here: Board decks, grant drafts, student communications, data analysis scripts, branded documents — anything that produces a file or touches an external tool.
This is your archive and version control.
Once a deliverable is ready for team review or final distribution, move it to your team's SharePoint folder. SharePoint provides co-authoring, comments, version history, and access controls. It is where finished documents get stored, shared externally, and maintained over time.
What goes here: Finalized documents, approved communications, board materials, anything that needs version history or external sharing.
Many tasks stay entirely within Projects or Cowork. You only move to SharePoint when a deliverable is ready for team review, external sharing, or archival.
When you open Cowork, it asks you to select a folder on your computer. This is your working folder — a dedicated space where Claude saves files during a session.
Create a folder like Claude_Working on your Desktop. Point Cowork at this folder every session. When a file is finished, move it to SharePoint or wherever it belongs. Keep the working folder clean.
Your working folder is a scratchpad. SharePoint is the archive. A Claude conversation is the log of how you got there.
Once you complete all three prerequisites and receive your Claude Enterprise credentials, here is what your first session looks like:
Try this in under 5 minutes:
The key insight: Chat and Cowork are where you do the work. SharePoint is where finished work lands.
This is CRITICAL operational content.
When Cowork asks you to select a folder, choose a dedicated working folder. This folder is where Claude saves files during a session. It is a workspace, a scratchpad.
Claude_Working on your desktop or in a temporary locationAfter Claude generates a deliverable, save it to the appropriate SharePoint subfolder. This is where version history lives, where your team collaborates, and where the final product is stored.
[Team SharePoint] > AI Drafts > [Project Name]
Example: Student Success > AI Drafts > Spring 2026 Student Comms
"Create an 'AI Drafts' subfolder in your team's SharePoint. This keeps AI-generated work organized and separate from finalized documents."
"A Claude conversation is a log of a working session. It is a record of how you got to a draft. It is not a document archive. If you need to find the latest version of something, go to SharePoint."
| Location | Purpose | Persistence |
|---|---|---|
| Cowork Working Folder | Active production | Temporary — clear regularly |
| SharePoint AI Drafts | Draft storage & collaboration | Permanent — version history |
| Claude Conversations | Session logs | Searchable but not archival |
Click any ring to explore responsibilities
Know the four-tier data classification framework:
Student PII, FERPA-protected data — NEVER upload to Claude
Internal financials, HR data, draft strategy — upload only with approval
Internal comms, project plans, process docs — OK for Claude with judgment
Published reports, website content, public data — always OK
"When in doubt, classify one tier more restrictive than you think."
"Every AI-generated output that leaves DC CAP must be reviewed by a human before it is sent."
"Claude drafted it. You own it."
"AI fluency includes knowing when to use AI and when to do the work yourself."
When Claude confidently states something that isn't true. Like when a person misremembers a fact. Always verify Claude's factual claims, especially names, dates, statistics, and specific details. This is why the review step matters.
How much information Claude can hold in one conversation. Think of it as how many pages Claude can read at once. DC CAP's Claude instances have a large context window, so you can include substantial documents without worrying about length.
Your question or instruction to Claude. Clearer, more specific prompts produce better outputs. "Summarize this document" is weaker than "Summarize this document focusing on key findings for a funder audience." Details matter.
Revising your request based on what Claude gives you. "Make this shorter." "Add an example." "Rewrite this in a more conversational tone." This is how you get good results. Iteration is where fluency develops.
Pre-built instructions that encode DC CAP's organizational knowledge. They activate automatically when you describe your task. Example: our "funder-framing" skill knows DC CAP's voice, evidence standards, and target audiences. You write what you need; the skill shapes how Claude responds.
A small unit of text Claude processes. You don't need to worry about this—Claude handles it automatically. Tokens are how we count language processing, but from your perspective, just think about the length and complexity of what you're asking.
Every piece of data has a classification. Check it before sharing with Claude. If it contains student names, grades, financial details, or any PII, it is Tier 1 Restricted and does not go into Claude. When in doubt, ask.
AI-generated output is a first draft. It requires human review for accuracy, tone, policy compliance, and audience fit. The review is the work. Skipping it is the risk.
When AI contributed meaningfully to a deliverable shared with external audiences (funders, partners, the public), note the AI's role. Internal work products do not require disclosure unless your team lead requests it.
The first output from Claude is rarely the best output. Push back. Refine. Ask Claude to try again with different constraints. Iteration is where fluency develops. Accepting the first draft is where bad habits form.
Your name goes on the output, not Claude's. You are accountable for factual accuracy, data framing, tone, and policy compliance of everything that leaves your desk, regardless of how it was produced.