Team Yaito Pmo

Team Yaito — A Multi-Agent PMO on Microsoft Copilot Cowork

> Microsoft Agent Academy Hackathon · Cowork Collective track

Team Yaito is a multi-agent Project Management Office (PMO) built entirely on Microsoft Copilot Cowork. A single orchestrator agent (Yaito) coordinates 19 specialist agents across five PMO functions to plan, govern, measure, and scale real enterprise program delivery — with a human accountable for every consequential action.

Demo video (5 min): https://youtu.be/Sri1vZOrUAw

---

The Problem

Enterprise PMOs run program delivery on manual, fragmented work — status decks, scattered chat threads, and one-off spreadsheets. Planning, governance, and reporting consume senior PM time, knowledge lives in people's heads, and leadership lacks a real-time, defensible view of delivery health.

At FinX Group, a four-company financial-services organization serving 4,000+ employees, a 30-day HR Chatbot migration ("AgentHR") needed PMO rigor that a single PM could not sustain alone.

Target user: Program and project managers in regulated enterprises who must plan, govern, measure, and scale complex delivery under tight deadlines — while keeping a human accountable for every decision.

---

The Solution — Team Yaito

An orchestrator (Yaito) decomposes each request and delegates to 19 specialist agents organized into five PMO functions. Agents act on live enterprise context through Cowork's tools (Outlook, Teams, SharePoint, Calendar) and route every write action through a human-in-the-loop approval dialog.

Architecture

                    ┌─────────────────────────┐
   PM (natural  →   │   Orchestrator: Yaito    │
   language)        └────────────┬────────────┘
                                 │ delegates
        ┌──────────┬──────────┬──┴───────┬──────────┐
        ▼          ▼          ▼          ▼          ▼
     PLAN      ARCHITECT    GOVERN     MEASURE     SCALE
   (scoping,  (solution    (forums,   (metrics,   (reuse,
    schedule,  design,      approvals, quality,    follow-on
    risk)      decisions)   gates)     reporting)  agents)
        └──────────┴──────────┴──────────┴──────────┘
                                 │  via MCP tools
                                 ▼
        Microsoft Graph: Outlook · Teams · SharePoint · Calendar
                                 │
                                 ▼
                  Human-in-the-loop approval dialog
                  (nothing executes without confirmation)

| Function | What its agents do | |----------|--------------------| | PLAN | Scoping, scheduling, risk & dependency tracking | | ARCHITECT | Solution design, technical decision records | | GOVERN | Governance forums, approvals, compliance gates | | MEASURE | Metrics, quality scoring, status reporting | | SCALE | Reuse, follow-on agent creation, knowledge capture |

Data flow: Request → orchestrator → specialist agent → MCP tool retrieves/acts on tenant data → result returns to orchestrator → human approves → action commits.

---

Key Technologies

• Microsoft Copilot Cowork — the runtime where all 19 agents live and execute
• Microsoft Graph API — how agents read and act on enterprise data across Outlook, Teams, SharePoint, and Calendar
• Microsoft 365 Copilot — grounding and reasoning over tenant context
• Model Context Protocol (MCP) — the tool interface connecting agents to Graph-backed services
• Markdown + YAML agent definitions — how each of the 19 agents is declared and configured

---

How to Run

Team Yaito runs entirely inside Microsoft Copilot Cowork — there is no separate app to build or deploy.

1. Open Microsoft Copilot Cowork with an M365 account that has Outlook, Teams, SharePoint, and Calendar access.
2. Load the Team Yaito agent set — the orchestrator (Yaito) plus 19 specialist agents, each defined as a Markdown + YAML file in this repository.
3. Place the agent definition files in your Cowork agents folder so the orchestrator can route to them.
4. Start a conversation with Yaito in natural language (e.g., "plan the AgentHR migration governance for this week"). Yaito decomposes the request and delegates to the right specialist agents.
5. Approve actions as they surface — every send, post, schedule, or create routes through a human-in-the-loop approval dialog before it executes.

> No code, build step, or API keys are required.

---

Results — AgentHR Proof Workload

On a real 30-day HR-chatbot migration toward a 15 June 2026 go-live across four operating companies, Team Yaito delivered:

| Metric | Result | |--------|--------| | PMO staffing | Compressed a five-person PMO to 1 PM + 19 agents | | SQL correctness | 0.58 → 0.94 (+62 points) | | Governance forums | 8 / 8 cleared | | Follow-on agents seeded | 6 | | Scope | 4,000+ employees · 4 operating companies |

---

Design Principles

• Orchestration over model cleverness — value comes from decomposing a real PMO into the right specialist roles, not from a single large prompt.
• Grounded, not generative — agents reason over retrieved tenant context (emails, files, meetings) rather than free-form output.
• Human-in-the-loop by design — no consequential action executes without an approval dialog, making the system safe to run against a production tenant.

---

Hackathon Submission

• Track: Cowork Collective
• Submission type: Individual
• Demo video: https://youtu.be/Sri1vZOrUAw

Organization details masked as "FinX Group" for confidentiality. No confidential, proprietary, or sensitive information is included in this submission.

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Jirawat-Yaito
• Source: Jirawat-Yaito/team-yaito-pmo
• License: MIT

Install and usage instructions live in the source repository linked above.