DC CAP Enterprise AI  ·  Week 1

Hand Claude real work.
Safely.

Working confidently with AI requires an understanding of how to work safely. This week we get to the heart of safely using AI in your daily work.

Week 1 of 8 About 30 min · at your pace Nothing to turn in
Week 1 · the setup

Here's how Week 1 goes

Last week you got a win and met the machine. This week you hand Claude something real from your own plate. Three short moves: pick up the one habit that keeps your work safe, sort what's safe to share, then hand over a real task and watch Claude do the drafting. At the end you'll reread the Governance Acknowledgment you signed in Week 0 — after this week's practice, you'll know exactly what you agreed to.

The one line we avoid: individually identifying scholars or families. We can use DC CAP information and situational context to design our work paths while protecting scholars' data.

That line is easy to see, and once you have the habit below, you stay on the right side of it without thinking about it. You are always the one who decides what goes in.

Move 1 · the habit

Describe the work, protect the data

Remember how Claude works: it builds an answer from what you put in and the patterns it learned. It doesn't need a scholar's record to draft a re-engagement email or sketch a check-in plan. It needs the situation. So you give it the situation in your own words, and the names and numbers stay in your files.

The habit: Describe your vision, context, and need, leaving sensitive data in secure files. Watch Claude work as you manage the process.

Say it back: "I describe the situation; the names and numbers stay in my files."

Two ways to ask

Same task — re-engage a scholar who's gone quiet. Watch the difference in what each one hands over.

Don't

"Maria G., GPA 2.1, SSN on file, hasn't logged in since Oct 3 — write her a re-engagement email."

Do

"I have a first-gen sophomore who's missed our last two check-ins and seems at risk of not re-enrolling. Draft a warm, low-pressure email inviting them to reconnect."

What's the difference?

Both get you a strong draft. Only one put a scholar's record into the chat.

Show why it matters →

The safe version gave Claude the situation, which is all it needed. Your read of the scholar's situation is what made the email work; the name and the GPA only added risk. That's the whole habit: you keep the judgment, Claude gets the situation, the record stays in your files.

Move 2 · the doing

Sort what's safe

Before anything goes into Claude, you check one thing: what kind of information is this? DC CAP groups every piece of information into four tiers by how sensitive it is. You don't need to memorize them; you need the instinct for the top one. Tier 1 never enters Claude. Try a few below; one is trickier than it looks. Pick a tier, then see the answer.

A scholar's file: name, GPA, SSN, family contacts.

Tier 1 · Restricted. Anything that points to one scholar or family stays out of Claude. Describe the situation instead.

This fall's retention rate across all scholars, no names.

Tier 3 · Internal. A rate across all scholars, with no one singled out, is aggregated and de-identified. Fair game for analysis and planning in Claude Enterprise.

The board's strategy deck and the FY27 OKRs.

Tier 3 · Internal. Strategy, plans, and goals are internal working material. Useful context for Claude, with no individual at risk.

The published annual impact report on dccap.org.

Tier 4 · Public. Already published for anyone to read. The easiest case.

Outcomes for the six scholars at one small partner campus, broken out by result.

Tier 2 · Sensitive. This is the one most people miss, and it's the whole point of the habit. Taking the names off doesn't make it safe. With only six scholars, anyone who knows that campus can work out who's who. When a group is small enough to re-identify (under ten), treat it as sensitive.

That last one is the habit's edge: removing a name isn't enough when the group is small enough to point back to a person. The full tier table, with examples for every case, lives on the Governance Framework; pull it up whenever you're unsure.

Your turn: glance at your own desk. Name one file you'd never put into Claude, and one you safely could. That instinct is the skill.

See the full data tiers →

Move 3 · the craft

Brief it like an intern

Back in Week 0 you pictured handing work to an eager intern, and asked what they'd need to know to help. Prompting Claude well is the same move: a good prompt is a good brief. Five parts turn a vague ask into one Claude can act on, so the first draft comes back close.

  • Role. Who Claude should be. "You're a DC CAP coach writing to a busy parent" sets the voice and the expertise.
  • Context. Your situation, described. This is the habit you already have: what's going on, in your own words, no names or records.
  • Task. Exactly what you want, and any limits; specific beats vague.
  • Example. Show one. One line in the tone you want shows Claude more than three sentences describing it.
  • Format. What done looks like: an email, three bullets, a table, under 150 words.

Then you go back and forth. The first draft is a starting point; tell Claude what to change and it adjusts.

Weak prompt → strong prompt
Weak

Write an email about the scholarship deadline.

Strong

You're a DC CAP coach [role]. A first-gen senior is coming up on a scholarship renewal deadline next week and is juggling a heavy term, the kind of moment where a warm nudge helps [context]. Draft a short, warm reminder that the renewal is due and offers help [task]. Match the tone of "Hey, just checking in, no stress, here's the one thing for this week" [example]. Friendly, under 150 words, one clear next step [format].

What changed?

Both ask for the same email, but the weak one makes Claude guess everything: which deadline, who it's for, how it should sound. The strong one briefs it like an intern, and the draft comes back far closer to what you need.

See the five parts →

Role (a DC CAP coach), Context (the kind of situation, described, never a specific scholar's record), Task (a warm reminder that offers help), Example (the tone line in quotes), Format (under 150 words, one next step). You won't label them in real life; you'll feel when one is missing, because that's where Claude starts guessing.

See it on something big. The same event-planning task, briefed at three depths:

Level 1 · VagueRCTEF
Your prompt

“Help me plan a scholar event.”

What you get

A generic checklist: date, venue, invites, food, thanks.

Claude knows nothing about DC CAP, the four campuses, or your coaches, so it can’t touch the hard part.

Level 2 · Some contextRCTEF
Your prompt

“We’re DC CAP. Plan a meetup for our scholars at American, Catholic, Trinity, and ASU Local.”

What you get

A phased plan: timeline, task list, a basic comms plan.

Useful, but it treats the four universities as a mailing list. No plan for coordinating with them as partners, no owners, no risks, no metrics.

Level 3 · Full five-part briefRCTEF
Your prompt

“You’re a DC CAP program manager [R]. Four partner campuses, coaches own scholars, a partnerships lead owns the relationships, leadership sponsors [C]. Design the full workflow: owners, an 8-week work-back, the four-school coordination model, comms, risks, metrics, Monday mapping [T]. Match this rigor [E]. As a Monday-ready plan [F].”

What you get
  • Owner map · 6 workstreams
  • One liaison per campus
  • 8-week work-back
  • Three comms tracks
  • Risks + success metrics
  • Monday board structure

Each of the five parts did real work: the coordination model, the owner map, the partnership layer, the Monday structure. A plan your team can run.

Same task, same Claude. Between Level 1 and Level 3, the only thing that changed was the brief.

Pocket check, before you send a prompt: Role? Context? Task? Example? Format? If Claude would have to guess at one, that's the one to add.

There's more when you're ready: giving Claude several worked examples to copy from, asking it to think step by step, organizing a long prompt. Those come in later weeks; the five-part brief is plenty to prompt well now.

Move 4 · the win

Hand over a real task

Now put it together on something already on your plate. Pick one real task (an email you owe, a plan you're shaping, a document to tighten) and brief it with the five parts, keeping the situation described and the record in your files. Handing the task off is delegation. Briefing it clearly is description. You'll feel both this week; they're two of the four core skills this whole program builds.

Watch, then try
YouOur fall retention is up two points across all scholars this year, and I want my team focused on fall re-enrollment next. Draft a short, warm update: name the win first, then point us at the next push. Keep it specific and encouraging.
ClaudeSubject: Fall retention is up, and here's where we go next

Team: real progress this fall. Retention is up two points across all our scholars, and that's your check-ins and your follow-through showing up in the numbers. The next push is fall re-enrollment, so let's get ahead of it this month. I'll bring the details Friday. For now, take the win. — [Your name]

Notice what went in: the situation and a number that names no one. A rate across all your scholars is aggregated and safe to share.

[Role: who Claude should be, e.g. a DC CAP coach.] [Context: your situation in general terms, no names or records.] [Task: exactly what you want, plus any limits.] [Example: one line showing the tone or shape you want.] [Format: what done looks like, e.g. an email under 150 words.]
Your turn:
  1. Click Copy above.
  2. Open Claude and click the message box.
  3. Paste it in, then fill the five brackets with your own task, in your own words.
  4. Send, read what comes back, and add one detail if it isn't quite right.
The more of the situation you give it, the more useful the draft. The record stays where it is the whole time.
Week 1 · close the loop

Reread what you signed

You signed the Governance Acknowledgment in Week 0 to turn on your Claude seat. Back then it was a set of rules to accept. Now you've practiced the tiers, caught the small-group trap, and handed Claude a real task the safe way. Open it again and it takes five minutes to confirm the rules you signed are the habits you already have. Haven't signed yet? Do it now — your seat stays off until you do.

Reread the Governance Acknowledgment

It covers the four tiers you just practiced, the one line we don't cross, and what to do in the rare case something slips in. Signed in Week 0? A reread is all this takes. Not signed? Check the box while you're there — that's what turns on your seat.

Open the Acknowledgment →

One thing worth knowing before you ever need it: if Tier 1 information ever lands in a chat by accident, you delete that conversation right away and tell the AI Governance Lead or Co-Owner within two hours. They take it from there (the full steps live on the Governance Framework). Mistakes happen; handling them quickly is the job.

Your win

What you can do now

You can hand Claude real work and keep every scholar's data exactly where it belongs. You'll carry that confidence into every week from here: you hand Claude real work, and you keep the judgment.

This is your own check that the week landed. It stays in your browser — nobody else sees it.

Now you know how to work safely with AI. Let's create something great!
What's nextWeek 2 · Use it on your real work
See the Weekly Map →
Optional

Under the hood

Skip it and lose nothing.  Here for the curious — why the safe way works as well as the unsafe way, and a couple of edges worth knowing.

Why does describing work as well as pasting the record?

Because Claude writes from the situation you describe and the patterns it learned about warm, clear writing. A name and a GPA add risk without adding quality. The one place a real detail helps is when it changes the substance — "they're a transfer student," "their financial aid is in question" — and you can give that as a description without ever naming the person.

Is removing the name enough?

Usually, but not always. If a group is large, an anonymized summary stays anonymous. If a group is small, say six scholars at one campus, the surrounding detail can point right back to a person even with no name attached. DC CAP's rule: when a cell is under ten and could identify someone, treat it as sensitive (Tier 2). When in doubt, describe the pattern rather than share the rows.

Why does it have to be Claude Enterprise?

DC CAP's Claude Enterprise workspace is set up so your inputs aren't used to train the model, there's an audit trail at the account level (not a log of your conversation contents), and data handling meets the program's governance standards. A personal AI account has none of that, and consumer tools may use what you type to train. So DC CAP work, at any tier, stays in the Enterprise workspace.