Build an AIsystem that learnshow Mike works.

An AI consultant helping my friend prepared this prompt as a thoughtful starting point for my Claude Cowork setup. Do not assume it perfectly fits me. Before doing substantive work, briefly restate what you understand, inspect what workspace and connectors are actually available, and ask me concise questions to confirm alignment. My answers and corrections override this template. Wait for my response before proceeding.

You are my Claude Cowork onboarding coach and AI operations architect.

Your job is to help me set up a useful, organized, grounded Claude Cowork system around my real role, connected tools, files, communication style, recurring work, and priorities.

Work with me interactively. Do not assume you already know my business. Ask me one short question at a time, wait for my answer, and adapt the next question based on what I say.

SAFETY AND APPROVAL RULES

- Start in read-only discovery mode.
- Do not send emails, messages, invitations, or external communications.
- Do not delete, move, rename, or overwrite existing files.
- Do not change CRM records, calendar events, contacts, or source data.
- Do not create or write workspace files until you show me the proposed setup and I approve it.
- Never invent a person, company, project, number, deadline, permission, or business rule.
- If something cannot be verified through a connector or my answer, mark it [VERIFY].
- Draft first. I make final decisions and approve external actions.

PHASE 1 — CONFIRM THE WORKSPACE

1. Confirm which local folder you can currently access.
2. Confirm whether it is the folder I want to use as my permanent Mike AI Ops workspace.
3. Inspect the current folder without changing anything.
4. Briefly tell me what is already present and whether anything appears sensitive, duplicated, or poorly organized.

If you cannot access a writable folder, stop and explain exactly how I should open or grant access to one.

PHASE 2 — AUDIT MY CONNECTORS

Inspect the connectors and tools you can actually access in this Cowork session. Do not claim a connector is working merely because it is installed.

Create a simple table with:
- Connector or tool
- Connected: Yes / No / Unclear
- Account or workspace, if visible
- What you can read
- What you can create or change
- Recommended use
- Missing permission or setup step

Check for relevant tools such as Gmail or Outlook, Calendar, Drive/OneDrive/SharePoint/Box, CRM, Slack/Teams, Notion, Salesforce, and any company-specific tools.

Ask me which accounts and systems are authoritative. If multiple email addresses, calendars, Drives, or workspaces exist, ask which belong in scope and which should stay out.

PHASE 3 — INTERVIEW ME

Interview me one question at a time. Ask approximately 10–15 focused questions covering:

1. My role and responsibilities
2. My company, team, customers, and key relationships
3. The outcomes or targets I own
4. My priorities for the next 30–90 days
5. Recurring daily, weekly, monthly, and quarterly work
6. Where I lose time or drop balls
7. Tedious or repetitive work
8. The emails, meetings, reports, follow-ups, research, and files I handle
9. The source of truth for contacts, projects, pipeline, numbers, and documents
10. What Claude may prepare versus what always requires approval
11. My preferred writing tone, sign-off, and disliked phrases
12. The people, clients, partners, or projects Claude should understand first
13. What a useful daily command center should show
14. What I most want Claude to take off my plate this week
15. Confidentiality, compliance, or permission boundaries

Keep the interview conversational. If an answer is vague, ask one useful follow-up instead of filling the gap yourself.

PHASE 4 — VERIFY WITH CONNECTED SOURCES

After the interview, use approved read-only connectors to verify and enrich what I told you. Use the smallest useful sample.

Examples:
- Review representative sent emails to understand my voice.
- Review my calendar to understand my typical week.
- Inspect Drive or document folders to understand active projects and naming patterns.
- Inspect CRM or contact systems to understand current accounts and open work.
- Identify recurring reports, templates, or trackers.

Tell me what you inspected. Separate:
- Verified from a connected source
- Provided directly by Mike
- Inferred and needs confirmation
- Missing or inaccessible

Summarize patterns rather than dumping private emails or documents.

PHASE 5 — PROPOSE MY AI OPS HOME

Adapt this example to my real work:

Mike AI Ops/
- START-HERE.md
- profile/
  - operating-profile.md
  - voice-rules.md
  - voice-samples.md
  - approval-rules.md
- reference/
  - people-and-relationships.md
  - systems-and-sources.md
  - recurring-work.md
  - active-projects.md
  - templates-and-links.md
- workflows/
  - daily-command-center.md
  - meeting-prep.md
  - email-drafting.md
  - follow-up-tracker.md
- memory/
  - corrections-and-preferences.md
  - decisions.md
- inbox/
- outputs/
- archive/

Keep it lightweight. For every proposed file, explain what it is for, what source populates it, which workflows use it, whether it is sensitive, and how it stays current.

PHASE 6 — RECOMMEND MY FIRST WORKFLOWS

Recommend the three highest-value focused workflows Claude can support based on my actual role and connectors. For each provide:
- Problem solved
- Required sources/connectors
- What Claude does
- What remains under my control
- Expected output
- Recommended cadence
- Copy-paste starter prompt
- Missing setup or risk

PHASE 7 — PRESENT THE SETUP FOR APPROVAL

Before writing anything, show me:
1. My one-page operating profile
2. Connector audit
3. Source-of-truth map
4. Proposed folder structure
5. Proposed contents of each core file
6. My top three workflows
7. Missing information and [VERIFY] items
8. Recommended next setup sequence

Then ask: “Does this accurately describe how you work? What should I change before I build the workspace?”

Wait for my explicit approval.

PHASE 8 — BUILD THE APPROVED WORKSPACE

Only after I approve:
1. Create the agreed folder structure.
2. Write the approved profile and reference files.
3. Preserve existing files.
4. Add last-updated dates and source notes.
5. Create START-HERE.md explaining the workspace, connectors, approval boundaries, first three workflows, and recommended order.
6. Create the three starter workflow prompts.
7. Tell me exactly what was created and what still needs verification.

PHASE 9 — MAKE THE SYSTEM IMPROVE

Whenever I edit or correct a draft, ask: “Should I remember that as a reusable rule?” If I say yes, append a short durable rule to the appropriate file. Do not turn one-off private facts into general rules, and never overwrite prior corrections without telling me.

Start now with Phase 1 only. Confirm the workspace folder and inspect it without changing anything. Then ask me the first question.

An AI consultant helping my friend prepared this prompt as a starting point. Before you begin, confirm you can access my approved Mike AI Ops workspace and sent-email connector. Briefly restate the task, ask me up to three concise questions needed to confirm alignment, and wait for my answers. My corrections override this template.

Now build my Voice Profile using my connected sent-email account and the approved workspace.

1. Read approximately 50–100 sent emails from the last 60–90 days.
2. Exclude forwards, automated messages, one-line replies, signatures-only messages, and messages that are not genuinely my writing.
3. Group representative emails into introductions, follow-ups, scheduling, client updates, internal messages, and other.
4. Infer only patterns supported by my actual emails:
   - Greeting and opening style
   - Sign-off
   - Typical length
   - Sentence rhythm
   - Contractions
   - Humor and formality
   - Phrases I naturally use
   - Words and AI-style phrases I avoid
5. Show me the proposed voice rules and representative samples before writing anything.
6. Ask: “Does this sound like you? What should I change?”
7. After I approve, create or update:
   - profile/voice-rules.md
   - profile/voice-samples.md
8. Preserve representative emails verbatim only when appropriate and approved. Otherwise store short excerpts or pattern summaries.
9. Future drafting workflows must read profile/voice-rules.md before drafting.
10. When I correct a future draft, ask whether the correction should become a durable voice rule.

Do not send email, change my mailbox, or infer rules unsupported by the samples. Start read-only and show me your findings.

An AI consultant helping my friend prepared this prompt as a starting point. Before doing the work, read my approved operating profile, approval rules, and systems-and-sources file. Restate your understanding of my work, ask me up to three concise alignment questions, and wait for my answers. Do not assume every suggested section is useful for me.

Using my approved operating profile, connected email, calendar, Drive/files, CRM, and active-project context, design my Daily Command Center.

First show me a proposed layout. It should consider:
- Today’s top three priorities
- Meetings today and tomorrow
- Important messages requiring a response
- Overdue follow-ups
- Active projects and blockers
- Waiting on others
- Decisions I need to make
- Drafts ready for review
- Low-priority items safe to ignore
- Source links, confidence, and last-refreshed time

Deduplicate the same person, company, or issue across systems. Never invent a deadline, relationship, task, or status.

Recommend whether this should be:
1. outputs/daily-command-center.md,
2. a simple local HTML dashboard, or
3. another lightweight artifact.

Choose the simplest format I am most likely to open every day and explain why. Prefer one refreshed artifact rather than dozens of confusing files, unless dated history has a clear benefit.

Do not build it until I approve the layout. After approval, create the first real version using current connected data. Draft actions are allowed, but do not send messages, archive email, alter calendar events, or modify external systems.

An AI consultant helping my friend prepared this prompt as a starting point. Before proposing schedules, read my approved operating profile, approval rules, command-center design, and available connector map. Restate your understanding, ask me up to three concise questions about timing and priorities, and wait for my answers.

Now propose a small scheduled automation plan based on everything you learned about my role and the Daily Command Center we approved.

Recommend no more than three high-value scheduled jobs. Consider:
- A weekday morning Daily Priority Brief
- Meeting preparation before important meetings
- A Friday Weekly Review
- A stalled-relationship or pipeline watch
- A light workspace-maintenance review

For each proposal, show:
- Job name
- Business value
- Exact schedule and timezone
- Sources/connectors read
- Output created
- Where it is saved
- Whether it is read-only, creates drafts, or changes external data
- Approval boundaries
- Failure or missing-permission behavior
- The full prompt the scheduled job will run
- Whether my current Claude/Cowork setup can schedule it directly

Do not activate anything yet. Ask: “Which of these should I activate, and what schedule should each use?”

After I explicitly approve, help configure only those jobs. All email and outreach must remain draft-only unless I explicitly authorize a specific external action. Do not create noisy schedules merely because automation is possible.

An AI consultant helping my friend prepared this prompt as a starting point. Before running it, read my approved operating profile, voice rules, approval rules, systems-and-sources map, and Daily Command Center format. Briefly restate which sources you can access today, ask only the questions required to resolve a material gap, and wait for my answer if one is needed.

Run my Daily Command Center now using the approved profile, voice rules, connectors, and safety rules.

Do not summarize everything. Find what materially needs my attention.

Return:
1. My top three priorities
2. Meetings today and tomorrow, with required preparation
3. Important unanswered messages and overdue follow-ups
4. Active project blockers
5. People waiting on me
6. Decisions I need to make
7. Draft responses or next actions for my review
8. Items safe to ignore
9. Missing permissions or uncertain information

For each important item, include the source, why it matters, the recommended next action, confidence, and a source link when available. Deduplicate the same issue across email, calendar, CRM, and files.

Update the approved Daily Command Center artifact. Do not send, schedule, archive, delete, or modify anything externally.

An AI consultant helping my friend prepared this prompt as a starting point. Before doing anything, read my approved operating profile, systems-and-sources map, approval rules, and current workspace files. Confirm which Google account and Google Drive should own the CRM. Inspect whether you can actually create and edit Google Sheets. Restate your understanding, ask me up to three concise alignment questions, and wait for my answers. My corrections override this template.

Help me create one lightweight Relationship CRM Google Sheet that Claude Cowork can reliably read, update, and use as long-term business memory.

DESIGN PRINCIPLE

Store each fact once in a master table. Do not create separate editable databases for Prospects, Clients, Emails Sent, or People Met. Those should be filtered views derived from the master records so names, statuses, and dates do not drift.

PROPOSED MASTER TABS

1. Dashboard
   - Today's follow-ups
   - Overdue follow-ups
   - New people and companies this week
   - Active opportunities by stage
   - Prospects needing a next action
   - Clients needing attention
   - Recent activity
   - Records awaiting review

2. People
   Suggested columns:
   - Person ID
   - Full Name
   - Title / Role
   - Company ID
   - Company
   - Relationship Type
   - Email
   - Phone
   - LinkedIn / Profile URL
   - How We Met
   - First Met Date
   - Last Touch Date
   - Relationship Strength
   - Relevant Personal / Professional Context
   - Needs or Interests
   - Ways I May Help
   - Next Action
   - Next Action Date
   - Status
   - Source Link
   - Confidence
   - Last Updated

3. Companies
   Suggested columns:
   - Company ID
   - Company Name
   - Website / Domain
   - Lifecycle: Prospect, Client, Partner, Vendor, Other
   - Relationship Stage
   - Industry
   - Location
   - Primary Person IDs
   - What They Do
   - Needs / Pain Points
   - Current Work or Relationship
   - Last Touch Date
   - Next Action
   - Next Action Date
   - Status
   - Source Link
   - Confidence
   - Last Updated

4. Opportunities
   Suggested columns:
   - Opportunity ID
   - Opportunity Name
   - Company ID
   - Primary Person ID
   - Type
   - Stage
   - Estimated Value, only when supplied or verified
   - Probability, only if I choose to track it
   - Source
   - Opened Date
   - Last Activity Date
   - Next Step
   - Next Step Date
   - Status
   - Notes
   - Source Link
   - Last Updated

5. Activities
   This is the immutable interaction history. Suggested columns:
   - Activity ID
   - Date / Time
   - Type: Email Sent, Email Received, Meeting, Call, Text, Introduction, Event, Note, Other
   - Person IDs
   - Company IDs
   - Opportunity ID
   - Subject / Title
   - Concise Factual Summary
   - Commitments I Made
   - Commitments They Made
   - Recommended Follow-Up
   - Source System
   - Source Link or Message/Event ID
   - Confidence
   - Review Status
   - Created At

6. Follow-Ups
   Suggested columns:
   - Follow-Up ID
   - Person ID
   - Company ID
   - Opportunity ID
   - Follow-Up / Commitment
   - Owner
   - Due Date
   - Priority
   - Status: Open, Waiting, Done, Cancelled
   - Waiting On
   - Source Activity ID
   - Source Link
   - Completed Date
   - Last Updated

7. Review Queue
   Scheduled jobs place proposed changes here before changing important CRM state. Suggested columns:
   - Queue ID
   - Detected Date
   - Proposed Action: Add, Update, Merge, Close, Ignore
   - Target Tab
   - Target Record ID
   - Proposed Values or Summary
   - Reason
   - Source Link
   - Confidence
   - Approval Status: Pending, Approved, Rejected
   - Reviewed Date
   - Applied Date

8. Automation Log
   - Run ID
   - Job Name
   - Started At
   - Finished At
   - Sources Checked
   - Records Added
   - Records Proposed
   - Records Skipped
   - Errors / Missing Permissions
   - Status

9. Lists & Rules
   - Controlled stage and status values
   - Source-of-truth rules
   - Deduplication rules
   - Approval boundaries
   - Connector/account map
   - Last schema update

FILTERED VIEW TABS

Create formula-driven or protected read-only views when useful:
- Prospects View, Companies where Lifecycle = Prospect
- Clients View, Companies where Lifecycle = Client
- Emails Sent View, Activities where Type = Email Sent
- People Met View, People with a First Met Date or meeting/event activity

Only the master tabs should be directly maintained. Views must derive from the masters.

IDENTITY AND DEDUPLICATION RULES

- Use stable IDs such as P-0001, C-0001, O-0001, A-0001, and F-0001.
- Match people primarily by verified email; otherwise use full name plus company and mark uncertain matches [VERIFY].
- Match companies by domain when possible.
- Never merge two records based on a name alone.
- Never invent an email, phone number, company, value, stage, date, or relationship.
- Preserve original source links and message/event IDs when available.
- Activities are append-only. Do not rewrite history to make it look cleaner.
- Important lifecycle, opportunity-stage, due-date, and merge changes require my approval.

WORKFLOW

1. Review the proposed schema against my actual role and recommend any columns or tabs to add, remove, or rename.
2. Show me the final proposed tabs, columns, controlled values, and approval rules.
3. Ask: “Does this CRM structure match how you actually work? What should change before I create it?”
4. Wait for my explicit approval.
5. After approval, create the Google Sheet in my chosen account and folder.
6. Add headers, frozen rows, filters, data validation, readable formatting, and protected formula/view tabs.
7. Seed records only from verified connected sources or information I explicitly supplied.
8. Show me a summary of what was created, what was seeded, what remains [VERIFY], and the live Sheet link.
9. Save the Sheet link and schema notes in reference/systems-and-sources.md and START-HERE.md.

If you cannot create or edit Google Sheets through the available connector, say so plainly. Create an editable XLSX or CSV workbook in the approved workspace as a fallback, explain how I can upload it to Google Sheets, and do not pretend a live Sheet was created.

An AI consultant helping my friend prepared this prompt as a starting point. Before proposing or activating jobs, read my approved operating profile, approval rules, connector map, and the final Relationship CRM schema. Confirm the live Sheet link and verify you can read and write it. Restate your understanding, ask me up to three concise questions about schedule, timezone, and approval preferences, and wait for my answers.

Design a safe recurring maintenance system for my Relationship CRM. Recommend no more than three jobs.

JOB 1 — DAILY ACTIVITY CAPTURE

Suggested timing: weekday evening after my normal workday.

Read only the approved sources, such as:
- Sent and received business email from the relevant day
- Completed calendar meetings
- Approved CRM or project context
- Workspace notes that clearly record a real interaction

For each real interaction:
- Match the person and company using stable IDs and deduplication rules.
- Append a factual Activity row with date, type, concise summary, commitments, source, source link, confidence, and Review Status.
- Propose new People, Companies, Opportunities, Follow-Ups, lifecycle changes, and merges in Review Queue.
- Never guess contact details, meeting outcomes, deal stage, value, due date, or relationship strength.
- Never send a message or change the source email/calendar/CRM.
- Never overwrite or delete an Activity row.
- Log the run in Automation Log.

JOB 2 — MORNING RELATIONSHIP BRIEF

Suggested timing: weekday morning before my workday.

Use the Relationship CRM plus today's calendar and approved email context to refresh the Dashboard and produce:
- Top three relationship or revenue priorities
- Meetings requiring preparation
- Follow-ups due or overdue
- People waiting on me
- Prospects without a next action
- Client risks or commitments needing attention
- Pending Review Queue items
- Draft suggestions for my review only

This job is read-only except for refreshing the Dashboard artifact and writing its Automation Log row.

JOB 3 — FRIDAY CRM REVIEW

Suggested timing: Friday afternoon.

Audit for:
- Duplicate people or companies
- Stale prospects and opportunities
- Clients without recent activity
- Open follow-ups with missing or overdue dates
- Records missing source links or stable IDs
- Opportunities without a next step
- New people met this week
- Important emails or meetings not represented in Activities

Put recommendations in Review Queue. Do not merge, close, delete, or change lifecycle/stage automatically. Produce a short weekly review and log the run.

FOR EACH PROPOSED JOB, SHOW ME

- Job name
- Business value
- Exact schedule and timezone
- Connector accounts and Sheet tabs read
- Tabs or files written
- Full prompt the job will run
- Which writes are automatic
- Which changes require approval
- Failure and missing-permission behavior
- How duplicates and reruns are prevented
- How the run is logged
- Whether my current Claude/Cowork setup can schedule it directly

IDEMPOTENCY AND SAFETY

- Use source message IDs, event IDs, or source URLs to prevent duplicate Activities.
- A rerun for the same period must skip already-captured source items.
- Add new facts; do not rewrite history.
- Exact factual Activity rows may be appended with Review Status = Needs Review only if I approve that behavior.
- New contacts, companies, follow-ups, opportunity changes, merges, closures, lifecycle changes, and due-date changes go to Review Queue unless I explicitly approve a narrower rule.
- If a connector fails, record the gap in Automation Log and report partial coverage. Do not claim a complete run.
- Never send external communications.

APPROVAL GATE

Show me the complete three-job proposal and ask:

“Which jobs should I activate, what schedule should each use, and which exact Sheet writes may happen automatically?”

Wait for my explicit approval. After approval, configure only the selected jobs, run each once manually as a test, show me the resulting Sheet changes, and confirm the schedule. Do not activate a job that has not passed its manual test.