🚀 Launch Control
A Product Launch Coordinator built with Microsoft Dataverse: from data model to agents to dashboard.
This repo is the companion to the Launch Control LinkedIn series by James Oleinik, Product Director for Microsoft Dataverse. Over 15 episodes (3/week for ~5 weeks), we build a complete product launch coordination system from scratch, and open-source every line of code.
The Problem
Engineering teams are automating code reviews, DRIs, and incident response. But up the stack? Executives still want status. Cross-team projects still mean passing spreadsheets around. What if we applied the same agentic thinking to project coordination?
The Solution: Three Layers
┌──────────────────────────────────────────────────────┐
│ LAYER 3: MANAGEMENT & OPERATIONS │
│ Power App Dashboard │ Python SDK/pandas │ CLI │
│ Admin Skills (agentic administration at scale) │
├──────────────────────────────────────────────────────┤
│ LAYER 2: INTELLIGENCE │
│ Business Skills │ Agents (Copilot Studio, Claude) │
│ Agent Flows │ Custom Actions │ BYO MCP Servers │
│ MCP Server (universal) │ CLI │ SDK (coding agents) │
├──────────────────────────────────────────────────────┤
│ LAYER 1: BUSINESS SYSTEM OF RECORD │
│ Dataverse Tables & Relationships │ Prompt Columns │
│ Virtual Entities (real-time, no data replication) │
│ Dataverse Intelligence (native M365 Copilot) │
└──────────────────────────────────────────────────────┘
Episode Guide
Full series index with links to each episode's README, preflight, and scripts: episodes/README.md.
| # | Episode | Hero capability |
|---|---|---|
| 1 | AI-Powered Data Modeling | Official Dataverse plugins for Copilot & Claude Code → first Dataverse tables |
| 2 | Your Playbook & Ingestion | Business skills + mapping-driven CLI ingestion |
| 3 | Promoting the Staging Layer | Python + pandas; staging → unified |
| 4 | Extending & Enforcing the Model | Virtual entities (custom GitHub Issues) + a server-side business rule the coding agent authors: guardrails every future agent must honor |
| 5 | Custom Tools | Custom API + two BYO MCP custom connectors registered with paconn |
| 6 | Cowork Plugin for Dataverse | Build & publish a Dataverse-aware Cowork (Teams) plugin |
| 7 | Microsoft Scout 🟡 | The Dataverse MCP face-lift (16 tools, search_data) driven by an always-on Scout Autopilot agent on a schedule |
| 8 | Roles & Reach | Four flat roles (Member / Owner / Viewer / Admin) over Eps 1-5 data + tools: same query, four lenses |
| 9 | Dataverse + F&O | CRM and ERP on one platform; the launch record gains budget and supply signal |
| 10 | Dataverse + Web IQ | Outside-in agent (new Copilot Studio builder); internal blockers fused with live web/news/CVE signal |
| 11 | Dataverse + Fabric IQ | Autonomous agent reasoning over the semantic baseline; is this launch a statistical outlier? |
| 12 | Dataverse + Foundry IQ | Code-first agent grounded in federated, cited knowledge (Foundry IQ over Azure AI Search) |
| 13 | Convergence | Native M365 Copilot (Dataverse intelligence / Work IQ) plus the three IQ agents, on one launch |
| 14 | Agentic Administration | The management plane is agent-driven: capacity, audit, cleanup, blast-radius |
| 15 | Full Orchestra + Your Turn | Six surfaces in 60 seconds + open-source CTA |
Each episode is also tagged in git: git checkout ep-08 to see the repo as it was at that episode's ship. The original Season 1 agent builds (declarative agent, autonomous Sentinel, code-first Python agent, generative dashboard) moved to episodes/archive/ and are absorbed as the Season 2 runtimes; see episodes/README.md for the full multi-season arc.
Quick Start
Prerequisites
- Python 3.10+ with
pip install PowerPlatform-Dataverse-Client - Node.js 18+ with
npm install -g @microsoft/dataverse - PAC CLI
- A Microsoft Dataverse environment with System Administrator role
Setup
git clone https://github.com/jamesoleinik/launch-control.git
cd launch-control
cp .env.example .env
# Edit .env with your Dataverse environment URL and credentials
pip install -r scripts/python/requirements.txt
To verify any episode is set up correctly, run its preflight:
python episodes/ep-05-custom-tools/preflight.py
Repo Structure
launch-control/
├── episodes/ # 15 per-episode folders + archive/ (READMEs + preflights + scripts)
├── datamodel/ # Table definitions, mappings, sample data
├── solutions/LaunchControl/ # Exported managed solution
├── business-skills/ # Launch readiness, escalation, status rules
├── data/knowledge/ # Sanitized KB articles for RAG
├── agents/
│ ├── launch-coordinator/ # Declarative Copilot Studio agent (archived; absorbed into Ep 10)
│ ├── launch-sentinel/ # Autonomous agent runtime (Ep 11 Fabric IQ)
│ ├── launch-coordinator-py/ # Code-first agent runtime (Ep 12 Foundry IQ)
│ └── agent-flows/ # Agent flow configurations
├── apps/launch-command-center/# Generative Power Apps page (folds into Ep 13 convergence)
├── plugins/ # Custom Dataverse plugins & actions
└── scripts/ # auth.py + shared Python utilities
Key Themes
Skills All the Way Down
- Build skills: AI creates your data model, plugins, and agents
- Process skills: Business knowledge agents follow at runtime (portable across agent platforms)
- Operate skills: Scripted automation and agentic administration at scale
Master the Process in Dataverse
The launch spans GitHub, SharePoint, email, but mastering the business process state in Dataverse (think dynamic programming) means every agent, app, and Copilot reads the same truth.
Contributing
See CONTRIBUTING.md for how to fork and adapt this for your own scenario: customer onboarding, release management, incident response, or anything else.
See also
CHANGELOG.md: what shipped in each episode (with git tags)SECURITY.md: how to report a security issueCODE_OF_CONDUCT.md: Microsoft Open Source Code of Conduct
License
Built by James Oleinik | Product Director, Microsoft Dataverse
No comments yet
Be the first to share your take.