architecture

Operations AI Agents

AI agents for operations: vendor communication, process documentation, and meeting summarization. Includes off-the-shelf options, custom build approaches, and training examples.

Overview

Operations roles handle process management, logistics, vendor coordination, and cross-functional support. Many tasks involve information synthesis and routine communication.

Efficiency Potential: 3-5 hrs/week at Adoptive tier

Agent Summary

Agent Efficiency Target Complexity Priority
70% faster drafting Low Medium
4 hrs → 1 hr Medium Medium
20 min → 5 min Low High

Vendor Communication Manager

Problem Statement

Operations manages multiple vendors with routine communications—status requests, issue escalation, renewal discussions. Each vendor requires context recall and professional communication.

Efficiency Target: Vendor communication drafting 70% faster

Training Example: Vendor Communication

Generate vendor communication based on context.

VENDOR: {{vendor_name}}
CONTACT: {{contact_name}}, {{contact_title}}
RELATIONSHIP CONTEXT: {{relationship_summary}}

COMMUNICATION TYPE: {{type}} (status request, issue escalation, renewal discussion)

CURRENT SITUATION:
{{situation_details}}

COMMUNICATION HISTORY:
{{recent_communications}}

OPEN ISSUES:
{{open_issues}}

OUR POSITION/ASK:
{{our_position}}

GENERATE:

Subject: {{appropriate_subject}}

[Professional communication that:]
- References relevant history
- States clear purpose
- Makes specific ask
- Sets appropriate urgency
- Maintains relationship

---

INTERNAL NOTES:
- Escalation path if no response: {{escalation}}
- Fallback position: {{fallback}}
- Timeline sensitivity: {{timeline}}

Architecture Pattern

Uses Pattern: Memory & State Management for vendor context:

# Vendor entity schema
{
    "vendor_name": "Acme Corp",
    "type": "vendor",
    "contacts": [
        {"name": "John Smith", "role": "Account Manager", "email": "..."}
    ],
    "contract": {
        "value": 50000,
        "renewal_date": "2026-06-01",
        "terms": "Annual"
    },
    "interactions": [
        {"date": "2026-01-02", "type": "email", "summary": "..."}
    ],
    "open_issues": [
        {"id": "ISS-123", "description": "...", "status": "open"}
    ],
    "relationship_health": "good"
}

Process Documentation Generator

Problem Statement

Operations owns process documentation that's perpetually outdated. Documenting processes is tedious; keeping them current is harder.

Efficiency Target: Process doc creation from 4 hrs → 1 hr; easier maintenance

Training Example: Process Document

Create process documentation from this information.

PROCESS NAME: {{process_name}}
PROCESS OWNER: {{owner}}
LAST UPDATED: {{date}}

INTERVIEW NOTES/DESCRIPTION:
{{process_description}}

SYSTEMS INVOLVED:
{{systems}}

STAKEHOLDERS:
{{stakeholders}}

GENERATE:

## {{process_name}}

**Owner:** {{owner}}
**Last Updated:** {{date}}
**Review Frequency:** [Quarterly/Annually]

### Purpose
[Why this process exists; what problem it solves]

### Scope
- **In scope:** [what's covered]
- **Out of scope:** [what's not covered]
- **Triggers:** [what initiates this process]

### Prerequisites
- [ ] [Required before starting]

### Process Steps

#### Step 1: {{step_name}}
**Who:** [Role]
**System:** [If applicable]
**Action:**
1. [Specific action]
2. [Specific action]

**Output:** [What's produced]
**Handoff:** [Who receives it next]

[Continue for all steps]

### Decision Points

| Decision | If Yes | If No |
|----------|--------|-------|
[Key decisions in the process]

### Exceptions & Escalation
- **Exception:** [scenario] → **Handle by:** [action]
- **Escalation path:** [when and to whom]

### SLAs & Metrics
| Metric | Target | Measurement |
|--------|--------|-------------|

### Related Documents
- [Links to related processes]

### Revision History
| Date | Change | Author |
|------|--------|--------|

Meeting Summarizer

Problem Statement

Operations participates in many cross-functional meetings. Notes are inconsistent, action items get lost, and follow-up is manual.

Efficiency Target: Meeting documentation from 20 min → 5 min; action items tracked automatically

Off-the-Shelf Options

Tool Capability Pricing
Otter.ai Transcription + summary $16/user/mo
Fireflies Meeting intelligence $19/user/mo
Fathom Free recording + notes Free
tl;dv Meeting recorder $25/user/mo

Training Example: Meeting Summary

Generate a meeting summary from this transcript/notes.

MEETING: {{meeting_name}}
DATE: {{date}}
ATTENDEES: {{attendees}}
PURPOSE: {{purpose}}

TRANSCRIPT/NOTES:
{{content}}

GENERATE:

## Meeting Summary: {{meeting_name}}
**{{date}}** | {{duration}}

### Attendees
[List with roles]

### Key Decisions
1. **Decision:** [What was decided]
   - **Context:** [Why]
   - **Owner:** [Who's accountable]

### Action Items
| Action | Owner | Due Date | Priority |
|--------|-------|----------|----------|
[Extracted from meeting]

### Discussion Summary
[3-5 bullet points of key topics discussed]

### Open Questions / Parking Lot
- [Items to address later]

### Next Meeting
- **Date:** [If scheduled]
- **Topics:** [Agenda items identified]

---
*Auto-generated summary. Review for accuracy.*

Implementation

# Integration with calendar for auto-triggering
class MeetingSummarizer:
    def __init__(self, transcription_service, llm):
        self.transcription = transcription_service
        self.llm = llm

    async def process_meeting(self, meeting_id: str):
        # Get transcript
        transcript = await self.transcription.get_transcript(meeting_id)

        # Get meeting metadata
        metadata = await self.calendar.get_meeting(meeting_id)

        # Generate summary
        summary = await self.llm.invoke(
            summary_prompt.format(
                meeting_name=metadata["title"],
                date=metadata["date"],
                attendees=metadata["attendees"],
                content=transcript
            )
        )

        # Extract and track action items
        actions = self.extract_actions(summary)
        await self.task_tracker.create_tasks(actions)

        # Distribute summary
        await self.email.send(
            to=metadata["attendees"],
            subject=f"Summary: {metadata['title']}",
            body=summary
        )