Welcome, Founders Cohort.

Your first ROCK prompt

Pick one real task from this week. ROCK is more than a prompt format: it keeps human purpose, responsibility, context, and earned judgment in charge of the work.

1. R · Role. Name your role in the work and the role you need AI to play. This defines the working relationship and the perspective AI should bring.
2. O · Objective. State the outcome you want from the beginning. When possible, attach an example of what good looks like. A clear destination keeps the work focused and gives you a standard for judging the result.
3. C · Context. Explain where you work, what is happening, what matters now, and the constraints around the task. Context gives AI the situation behind the request.
4. K · Know-how. Add the experience, standards, intuition, and judgment only you can provide. This may come from education, lived experience, prior roles, current responsibilities, or years in a profession. AI cannot infer it reliably on its own.

Role: I am [your role and responsibility in this work]. You are [the seat you need at your table, e.g., an experienced senior living operations analyst who writes plainly].

Objective: [One outcome, named. e.g., Research X and brief me in one page. / Analyze this spreadsheet and tell me the three things that matter. / Draft the memo for Monday's leadership meeting.] I have attached [an example of what good looks like, if available].

Context: I work at [organization type and size/scope]. Here is what is happening: [the situation, decision, or change]. What matters here: [2-4 facts the AI needs]. Do not use any resident or health information.

Know-how: Here is what my education, experience, professional judgment, or time in this work tells me matters, which you cannot know on your own: [2-3 rules of thumb or standards, e.g., "occupancy alone misleads; read move-outs against acuity mix"]. What good looks like to me: [your standard for done].

Working instructions: ask me up to three clarifying questions before you begin, then produce [format: a one-page memo / a table / an email draft]. Plain language. Keep it under [length].

Prompt library

A prompt library is the simplest documentation of a repeatable system: your personal record of instructions that have already produced useful work. Its value comes from being built by the person who uses it, not from collecting hundreds of generic prompts. It saves you from rediscovering what works and makes a good prompt easy to reuse, adapt, or share with a colleague.

• When to save one: after you have tested and corrected it, and expect to use it again or share it.
• What to keep: a clear name, what the prompt helps you do, when to use it, and the tested prompt itself. A date or testing note is optional.
• Where it goes: begin wherever you will reliably find it again. When the prompt becomes systematic, repeatable, or shareable, keep a reviewed Markdown file as the canonical record; create DOCX or PDF editions when they are easier for people to use.
• How to reuse it: copy it, adapt the details to the current situation, run it, and improve the saved version when you learn something.

Prompt name: [a short name you will recognize]

What it helps me do: [the useful outcome]

Use it when: [the situation where it applies]

Prompt: [paste the tested prompt here]

Notes from testing: [optional correction, variation, or reminder]

Created or updated: [optional date]

Skills

A skill is a reusable set of instructions for work you expect to do again. It captures the purpose, professional know-how, steps, standards, examples, and boundaries that make the work yours.

• When to create one: when a task repeats, depends on judgment, or should be performed consistently by you, a colleague, or an AI assistant.
• What it preserves: what triggers the work, the outcome, your rules of thumb, the steps, what good looks like, and what must never happen.
• How it is used: give the file to ChatGPT, Microsoft Copilot, Claude, or a colleague as the standing instructions for that task.
• How it improves: test it on real work, correct what is missing, add examples, and save a new version.

Why Markdown? Markdown is plain text with visible structure. It is easy for people to read and reliable for AI to follow. It turns thinking worth keeping into portable, durable documents that can move between people, applications, and systems.

You do not need to write Markdown syntax yourself. Describe the work in ordinary language, let AI interview you, and ask it to produce the finished Markdown file.

I want to create a reusable skill for [name the task].

Interview me one question at a time. Ask what you need to understand:
- when this skill should be used
- the outcome it should produce
- the professional know-how and rules of thumb I apply
- the steps I follow
- what good work looks like
- the boundaries and mistakes to avoid
- any example I can provide

Do not fill gaps with assumptions. After the interview, write the complete skill in clear Markdown. Show it in a Markdown code block and, if possible, provide it as a downloadable .md file. I will review and correct it before I use or share it.

Digital twin

A digital twin is a custom AI assistant configured with a description of your role, responsibilities, priorities, communication style, preferences, and working judgment. It is not a replacement or an impersonation. It is a working model that you review, correct, and control.

1. Ask what the tool already knows. ChatGPT can draw from prior conversations and memory when enabled. Microsoft 365 Copilot can draw from work context you are permitted to access through Microsoft Graph.
2. Review and correct it. Separate evidence from inference. Fix errors and add what is missing about your responsibilities, priorities, decisions, and communication.
3. Create the system instruction. Ask the tool to turn the reviewed description into clear Markdown instructions for how an assistant should work with you.
4. Configure and test it. Add the system instruction to a custom GPT, Copilot agent, ChatGPT Project, or Claude Project. Give it representative work and correct the instruction when needed.

Tell me what you know about me from the information available to you in this account or approved work context.

Organize what you know into:
- my role and responsibilities
- my current priorities
- how I make decisions
- how I communicate
- my working preferences
- recurring work and relationships that matter

Distinguish direct evidence from inference. Do not invent facts.

Then interview me one question at a time so I can correct errors and fill important gaps. When I approve the description, turn it into a system instruction for a digital twin that helps me think, communicate, and work.

Write the finished system instruction in clear Markdown. Include its purpose, what it knows about me, how it should work with me, my standards, its boundaries, and when it must ask rather than assume. Show it in a Markdown code block and, if possible, provide it as a downloadable .md file.

Digital chief of staff

A digital chief of staff is a bounded AI workflow for recurring coordination. It reviews the mail, calendar, files, and priorities you permit it to use, then prepares a briefing for your review. It supports the coordination work; it does not replace a human chief of staff or make consequential decisions.

1. Define its access. Use only approved accounts, permitted mail and calendar context, and information allowed by your organization's policy.
2. Teach your priorities. Give it your digital twin instruction and the triage standards, rules, and boundaries captured in your skill.
3. Test before scheduling. Run one briefing, inspect what it included or missed, and correct the instructions.
4. Put it on a schedule. After it works, schedule the briefing where your account permits. A tested scheduled assistant is your first agentic workflow.

Role: You are my digital chief of staff.

Objective: Prepare a concise daily briefing that helps me begin with the right priorities, commitments, and risks.

Context: Use only the mail, calendar, files, and other work context this approved account permits you to access. Apply my reviewed digital twin system instruction and the triage skill I created today. If required context is unavailable, tell me what is missing rather than guessing.

Know-how: Apply my reviewed priorities, triage rules, communication standards, and judgment captured in the digital twin and skill. When those sources do not answer a question, ask me rather than inventing a preference.

Produce exactly five sections:
1. Today's three priorities, with the reason each matters.
2. Meetings that require preparation, with one preparation action each.
3. Messages or commitments that need follow-up.
4. One risk, conflict, or unanswered question I may be missing.
5. One task I should defer, delegate, or stop.

Boundaries: Do not send messages, delete anything, accept meetings, make commitments, or take consequential action. Draft and recommend; I review and decide. Identify every assumption. Keep the briefing under 300 words.

Run it once now. Ask me what you got wrong or missed, then revise the instruction. After I approve it, help me schedule it for [days and time] where this account permits.

Shared agents for work

This is our neutral course term, not one vendor's product name. A shared agent for work is a reusable AI configuration for defined work, built from instructions, approved knowledge, and permitted tools or actions, then tested, governed, and shared within an organization. It may answer questions, create outputs, or take approved actions. How independently it operates depends on the product and configuration.

1. Choose one repeatable process. Define the user, outcome, and point where human judgment is required.
2. Add instructions and approved knowledge. Start with a reviewed skill, policy, template, or reference file.
3. Limit tools and actions. Grant only the access the process requires and require approval for consequential or write actions.
4. Test before sharing. Use synthetic or public data, inspect errors and assumptions, then identify the owner, reviewer, and escalation path.

Availability and capabilities vary by plan, license, and administrator settings. Shared agents should use approved data, least-privilege access, explicit review, and accountable ownership.

Visualize your work as a web page

Take a real Markdown file from the day, or a short set of notes in chat, and turn it into one calm HTML page you would be comfortable showing a colleague.

Take this Markdown file, or these notes from chat, and turn them into a single, clean HTML page I can open in my browser.

I want one page that is calm, readable, and built around the big ideas first.

Use the content below:

[Paste your Markdown file, prompt library note, skill draft, or other reusable working notes.]

Explore a topic through free-style conversation

Choose something you genuinely want to understand. Begin naturally, answer questions, challenge assumptions, and follow the most useful line of thought.

I want to explore [topic]. Ask what I already know and what I most want to understand. Then help me learn through questions, explanations, examples, and alternative perspectives. Correct me when needed, identify your assumptions, and help me decide what to explore or do next.

Turn everyday work into a reusable skill

Describe work you perform repeatedly. Ask AI to identify the separate skills inside it, choose one, and have the system build it for you.

Every day, I do this kind of work: [describe the work].

Ask me enough questions to understand it. Show me the reusable AI skills this work could be broken into and recommend which one to create first. After I choose, build that skill for me. Include its purpose, needed inputs, steps, quality checks, boundaries, and an example.

Create your digital twin

A digital twin is a personal assistant configured to understand how you work, communicate, decide, and set priorities.

Help me create my digital twin. Interview me about my responsibilities, priorities, audiences, communication style, standards, and boundaries. Ask one question at a time. Then create the assistant for me if this system supports it. Otherwise, give me the finished instructions and context file so I can reuse them.

Create your digital chief of staff

A digital chief of staff prepares a useful briefing or coordination product for you on a schedule, using only the information and access you approve.

Help me create a digital chief of staff that prepares [briefing or recurring output] every [schedule]. Ask what it should review, what it should produce, what information it may use, what it must never do, and what I must review. Then create and schedule it if this system supports that. Otherwise, give me the reusable prompt and a manual checklist.

Choose a use case and build its skill or agent

Open the use-case library, choose work that matters to you, and build the smallest useful version. Your private workshop email offers an individualized starting point, but you may choose any relevant use case.

Open the Senior Living AI Use-Case Library (PDF) · DOCX · Markdown

I chose this use case: [name or describe it].

Help me decide the smallest useful first build: a reusable skill, a personal assistant, a shared agent, or a scheduled prompt. Ask about the outcome, users, approved inputs, steps, professional judgment, quality checks, boundaries, and human review.

Then build the skill or agent with me using public, synthetic, or approved information. If this system cannot create it directly, produce the complete portable instructions and a simple test plan.

Make it portable, testable, and teachable

Preserve what worked so you can reuse it in another system and a teammate can understand, test, and improve it.

Help me turn what I built into a portable team package. Include:
1. The useful output or capability.
2. The reusable prompt, skill, or agent instructions.
3. The approved context and example another person needs.
4. The quality, privacy, and human-review checks.
5. A short test plan and a five-minute teach-back.
6. The next step toward a shared, leadership-approved workflow.

Produce the package in clear Markdown. Then help me test the same approved instructions in a second AI system and identify what must change.

Claude across the places we work

Claude is appearing across several work surfaces rather than only in a stand-alone chat. This is an introduction and group demonstration; participants do not need a Claude account or any of these products installed.

Group demonstration: provide one approved portable Markdown instruction to Claude.ai, then use a synthetic workbook in Claude for Excel to locate a formula, question an assumption, cite the supporting cells, and explain the finding in plain language.

The point is not that everyone should adopt every surface. The point is that AI assistance is moving into the browser, office applications, desktop, and technical tools where work already happens. Availability varies by plan, operating system, and administrator settings. Use only public, synthetic, or organizationally approved information.

Safe first experiment: start with trusted or synthetic files. Review every change, limit connected folders and network access, and follow Anthropic’s prompt-injection guidance before using an external spreadsheet.

Read Anthropic's Claude for Excel and prompt-injection guide

Leadership discussion: govern, scale, and measure AI

This is a live discussion with the full cohort. We will connect what participants built during the week to the organizational decisions required for responsible, durable adoption.

Direct library links: policy starter kit · governance and security settings · official product guides · use-case library · managing AI spend · reading list · public evidence

Governance and policy: start with data governance and security settings to see the controls behind access, retention, and training.

Risk and security: use the same shelf, then check the Claude for Excel guidance if spreadsheet prompt injection is the concern.

Systems and agents: open official product guides for the platform docs and the use-case library for workflows to adapt.

ROI and operating value: open managing AI spend for budgets, limits, and the cost model behind adoption.

Leadership and culture: open the reading list and public evidence for the ideas and examples that shaped the adoption story.

If you do not have an AI policy yet: use Policy Starter Kit to draft a first policy, print a checklist, choose champions, and keep Markdown versions under control as the team improves the work.

Discussion outcome: identify what your organization should build, teach, govern, and measure next—and what leadership must decide before broader deployment.

This is a cultural change, executed through technology

The CEO and leadership team remain accountable for the purpose, culture, priorities, resources, and results. Technology, people, operations, data, security, legal, finance, and clinical leaders each own part of the execution. No single title is automatically qualified to lead the whole change; the operating model matters more than the title.

When nobody is fully qualified: do not wait for a perfect candidate or crown one person the expert. Name an accountable sponsor, assemble people with complementary judgment, use external help for genuine gaps, and develop internal capability through the work.

In a smaller organization: the CEO sponsors the change; one credible leader serves as the named AI lead alongside existing responsibilities; an internal or external technology and security resource checks the controls; a small group of champions tests and teaches useful work.

In a larger organization: the CEO or executive committee sponsors the change; an operating owner coordinates the portfolio; a cross-functional council joins the required disciplines; business-unit champions carry adoption into real workflows.

Chief AI Officer: useful when the work requires a full-time enterprise owner with authority, budget, and cross-functional reach. The person may come from technology, operations, strategy, the liberal arts, or another field, but must connect people, business judgment, technology, data, governance, and measurable value. The title alone does not create capability.

Draft your first AI policy and operating model

If your organization has not written an AI policy yet, this is the first practical step. Use a prompt to generate a draft, a checklist to verify the basics, and a short champion-and-version note so the work can be shared without losing the current version.

Markdown is the operating file. A champion can circulate a tested copy, but the source version should stay named, dated, and easy to compare so the team knows what changed and why.

This is how policy becomes an operating model: leadership names the rules, the checklist verifies readiness, champions carry the method into real work, and Markdown preserves the best version for the next revision.

Leadership first: from personal know-how to organizational intelligence

We begin with individual leaders, including the CEO and leadership team, before asking the organization to follow. Leaders cannot responsibly sponsor, explain, govern, or fund work they have never experienced. Personal capability is the starting point, not the finish line.

1. Build firsthand fluency. Apply AI to work the leader understands deeply.
2. Use ROCK. Make Role, Objective, Context, and Know-how explicit.
3. Test the method. Correct the work and define the human judgment and boundaries.
4. Preserve what works. Save it in Markdown as durable, portable intelligence.
5. Teach through champions. One person teaches one useful build to another person.
6. Govern and measure. Add ownership, controls, evidence, and a decision about expansion.

The ripple: individual judgment becomes a tested method; the method becomes a portable record; a champion makes it teachable; the team improves it; the organization governs and measures the repeatable practice.

Choose champions for judgment and trust. Look for people respected for their existing work who understand a real workflow, verify results, teach clearly, follow boundaries, report failures, and have time and manager support. Include constructive skeptics as well as early enthusiasts.

Know your data before tools touch it

Data readiness means knowing what data you have, where it lives, who owns it, who may use it, and whether it is reliable enough for the proposed work.

• System of record: identify the authoritative source and data owner.
• Access: name who may see, change, connect, or export the information.
• Fitness: determine whether the information is current, complete, and reliable enough for this use.
• Boundaries: state what the AI must never receive, combine, conclude, or infer.

The data does not have to be perfect. It must be fit for the specific use, protected appropriately, and good enough to support a result people can trust. Connectors should come after these answers, not before them.

Draft a 90-day AI roadmap

Define what to build, teach, govern, and measure.

# My 90-Day AI Roadmap
Name: [you] · Written: July 17, 2026

## Leadership and ownership
Executive sponsor: [who holds enterprise accountability]
Operating owner: [who coordinates the work and evidence]
Champion: [who tests, teaches, and reports what happens]

## ROCK starting point
Role: [my responsibility in this work and the role AI should play]
Objective: [one 90-day operating outcome]
Context: [team, workflow, constraints, approved information, and systems involved]
Know-how: [the standards, judgment, experience, and rules of thumb we must preserve]

## What I build next (weeks 1-2)
[The smallest responsible assistant, skill, agent, or decision product; the build/partner/buy choice; and the human review.]

## What I show my team (weeks 2-4)
[Who sees it first. What they build with you. Teach one person one build within seven days and confirm they can repeat it safely.]

## Data readiness and governance
[System of record, data owner, fitness for use, approved and prohibited data, permissions, human review, tests, failure reporting, and stop condition.]

## How I measure
[Baseline and targets for adoption, quality, time redirected, economics, and risk. Retake the Leadership Index in October.]

## Durable record
[Save the reviewed roadmap, reusable instructions, examples, tests, and decisions in Markdown. Create DOCX or PDF editions when useful for human review.]

Reserve your bonus office hour

Before you leave: two brief surveys

Please complete both surveys now, while today’s roadmap work is still fresh. They help us understand where you started and what changed by the end of the sprint.

Leadership survey: Take the Leadership survey

Argentum survey: Take the Argentum survey

If you need to leave early, please try to finish both before you go.