Email Automation

Advanced

"Process last 50 emails"
"Check emails from last 3 days"
"Draft responses only"

How It Works

CRITICAL: Messages vs Threads

The Gmail API returns individual messages, not threads. A 200-message inbox may only be ~70 unique threads. Always deduplicate by thread_id before presenting results or counting:

from collections import defaultdict
threads = defaultdict(list)
for e in emails:
    threads[e['thread_id']].append(e)
# Use threads dict for counting and display

Step 1: Fetch Emails

• Connects to Gmail API via google-master integration
• Fetches inbox emails (default: all inbox, max 100)
• Extracts: sender, subject, snippet, full body
• Deduplicates by thread_id to get unique conversations

Step 2: AI-Powered Triage

For each email, AI analyzes:

• Sender importance (client, team, investor, recruiter, newsletter)
• Urgency signals (deadline mentions, "URGENT", meeting requests)
• Action required (question asked, decision needed, FYI only)
• Context (relates to projects, hiring, sales, operations)

Step 3: Generate Recommendations

• Draft response if reply needed (using your tone and context)
• Suggest archive if no action needed
• Suggest snooze with date (e.g., "snooze until Monday")
• Create Notion task if actionable work item

Step 4: Present for Approval

Shows summary table:

EMAIL TRIAGE RESULTS (25 emails processed)

│ # │ FROM              │ SUBJECT                    │ ACTION       │ DRAFT?  │
├───┼───────────────────┼────────────────────────────┼──────────────┼─────────┤
│ 1 │ [email protected]    │ RPO Pipeline Update        │ RESPONSE     │ ✅ Ready │
│ 2 │ linkedin          │ Job recommendations        │ ARCHIVE      │         │
│ 3 │ [email protected]   │ Q1 metrics check-in        │ SNOOZE (Mon) │         │

User can then:

• Approve all actions
• Review/edit specific drafts
• Skip certain emails

Configuration

Email Classification Rules

URGENT (needs immediate response within 24h):

• Client escalation ("issue", "problem", "not working")
• Investor asking for metrics
• Team member blocked ("need your input", "waiting on")
• Meeting today/tomorrow

RESPONSE NEEDED (reply within 2-3 days):

• Direct questions to you
• Introduction requests
• Partnership discussions
• Interview scheduling

TASK (create Notion task):

• "Can you review..."
• "Need feedback on..."
• "When you have time..."
• Long-form work items

ARCHIVE (no action needed):

• Newsletters
• Automated notifications
• CC'd for FYI
• Already handled by someone else

SNOOZE (defer):

• Non-urgent but important
• "End of week" / "next month" mentions
• Events/deadlines in future

Response Draft Style

Drafts use your communication style:

• Concise - 2-4 sentences max
• Direct - Clear yes/no, specific next steps
• Professional but friendly - Warm but efficient
• Action-oriented - Always include next step or deadline

Example draft:

Hi Sarah,

Great update on the RPO pipeline. The 3 new leads look promising.

Let's sync Friday 2pm to review the Zurich proposal. I'll send a calendar invite.

Jonas

Workflow

Full Inbox Triage

1. Run: "Process my emails"
2. Wait ~30-60 seconds (AI processes each email)
3. Review: Table with recommendations
4. Approve: "Execute all" or review individually
5. Done: Drafts saved, emails archived/snoozed

Draft-Only Mode

1. Run: "Draft responses only"
2. AI generates drafts for emails needing replies
3. Review: Drafts shown one by one
4. Edit/approve each draft
5. Done: Drafts saved to Gmail drafts folder

Quick Scan Mode

1. Run: "Quick email scan"
2. AI shows summary only (no drafts)
3. Highlights urgent items
4. Done: You decide what to action

Integration with Notion

Task Creation:

• Emails marked as TASK → Create entry in Tasks JBD database
• Properties set:
  • Action Zone: "Inbox Processing"
  • Status: "📥 Incoming"
  • Priority: Based on email urgency
  • Due date: Extracted from email if mentioned

Project Linking:

• If email relates to active project (detected by keywords), link task to project

Privacy & Security

• Read-only by default: Only reads emails, doesn't send without approval
• Draft folder: All responses saved as Gmail drafts for review
• No auto-send: You must approve before any email is sent
• Sensitive emails: Can flag certain senders as "never auto-draft" (investors, board, personal)

Superhuman Compatibility

Snoozed Emails

Superhuman uses a custom Gmail label for snoozes — not Gmail's native snooze:

• Label: [Superhuman]/Is Snoozed (Gmail ID: Label_3)
• Query: list_emails(label_ids=['Label_3']) — do NOT use in:snoozed

Superhuman AI Categories

Superhuman auto-categorizes emails with labels:

• [Superhuman]/AI/Respond (Label_29), AI/Waiting (Label_23), AI/Meeting (Label_24)
• AI/News (Label_25), AI/Investors (Label_30), AI/AutoArchived (Label_32)

Creating Drafts — KNOWN LIMITATION

Superhuman does NOT show drafts created via the Gmail API. Superhuman has its own draft system that doesn't sync inbound from Gmail.

Practical workflow:

1. Create threaded draft via create_threaded_draft() — appears in Gmail web (mail.google.com)
2. Or show the draft text directly to the user so they can copy-paste into Superhuman's composer
3. Always use threaded drafts (with threadId, In-Reply-To, References) for proper Gmail web threading

from gmail_operations import create_threaded_draft
create_threaded_draft(
    original_message_id='<msg-id>',
    to='[email protected]',
    subject='Re: Subject',
    body='Reply here'
)

Bulk Archive Workflow

For cleaning up large inboxes, use this pattern:

1. Fetch all inbox messages: list_emails(query='in:inbox', max_results=200)
2. Group by thread_id to get unique conversations
3. Categorize threads by keywords/sender patterns
4. Archive all messages in a thread: archive_email(msg_id) for each message with INBOX label
5. Save a tracker file before archiving (for reference)

Proven Category Keywords (from real triage)

• Investor/VC: ventures, capital, partners, fund, intro, catch-up, reconnect, beam <>
• Notifications: newsletter, no-reply, noreply, notification, amie
• Superhuman reminders: [email protected] (these are unsnoozed emails)

Advanced Features (Future)

• Learning your preferences: Over time, learns which emails you always archive vs respond to
• Response templates: Common responses ("Thanks, will review", "Let's schedule a call")
• Thread context: Analyzes full email thread, not just last message
• Calendar integration: When drafting meeting responses, checks your availability
• Bulk operations: "Archive all newsletters", "Snooze all recruiting emails to next month"

Example Session

You: "Process my emails"

AI: Fetching unread emails from Gmail...
    Found 23 unread emails from last 24 hours.
    Analyzing with AI...
    
    EMAIL TRIAGE RESULTS
    
    🔴 URGENT (2 emails)
    ├─ Christian Engnath - "When can we start?" → DRAFT READY
    └─ Fred - "Exec meeting agenda approval needed" → DRAFT READY
    
    💬 RESPONSE NEEDED (5 emails)
    ├─ Sarah - "RPO pipeline update" → DRAFT READY
    ├─ Giuseppe - "Second interview availability" → DRAFT READY
    └─ ... (3 more)
    
    📥 ARCHIVE (12 emails)
    ├─ LinkedIn - "Job recommendations"
    ├─ GitHub - "Dependabot security alert"
    └─ ... (10 more)
    
    ⏰ SNOOZE (4 emails)
    ├─ Investor - "Q1 metrics" → Snooze to Monday
    └─ ... (3 more)
    
    Actions:
    1. Review urgent drafts first? (say "show urgent")
    2. Execute all recommendations? (say "execute all")
    3. Review each email? (say "review each")

You: "Show urgent"

AI: [Shows draft for Christian]
    ---
    Hi Christian,
    
    Great to hear you're ready to move forward! 
    
    Can you start Monday Feb 17? Fred will send you the onboarding details today.
    
    Looking forward to working together.
    
    Jonas
    ---
    
    Approve this draft? (yes/edit/skip)

You: "Yes"

AI: ✅ Draft saved to Gmail drafts
    [Shows next urgent draft...]

Technical Implementation

Gmail API Endpoints Used

• users.messages.list - Fetch unread emails
• users.messages.get - Get full email content
• users.drafts.create - Save response drafts
• users.messages.modify - Archive emails
• users.threads.modify - Mark as read/snooze

AI Model

• Classification: GPT-4o-mini (fast, cost-effective for triage)
• Draft generation: Claude Sonnet 4.5 (better writing quality)

Caching Strategy

• Cache email classifications for 24 hours (avoid re-processing)
• Cache common response templates

Cost Estimation

• Per email: ~$0.002 (classification + draft if needed)
• 50 emails/day: ~$0.10/day = ~$3/month
• High volume (200 emails/day): ~$0.40/day = ~$12/month

Success Metrics

Track these to measure effectiveness:

• Time saved: Before/after time spent on email
• Inbox zero frequency: How often you reach zero unread
• Draft acceptance rate: % of AI drafts you send with minimal edits
• Archive accuracy: % of auto-archived emails that didn't need action

Status: Ready to implement
Priority: High (Phase 2.1 of CEO Office Automation)
Dependencies: Gmail API access (via google-master), Notion API (via notion-master)