How It Works
The Agentic AI Partner Library is a curated, searchable resource hub for the Agentic AI partner vTeam community. Every link in the library has been vetted by a domain expert — but the heavy lifting is done by AI agents.
What makes this different?
This isn't a static bookmark list. It's an agentic curation pipeline — AI discovers content, humans make decisions, and AI handles everything else. The library stays fresh, consistent, and growing with minimal manual effort.
The Curation Pipeline
Content flows through a fully automated pipeline with human judgment at the decision point.
1. Content Discovery
New content is discovered in two ways — both produce a structured GitHub Issue:
| Channel | How it works |
|---|---|
| 🤖 Research Agent | Scans 14 sources daily — RSS feeds, GitHub releases, Microsoft What's New pages. Filters by keywords and AI relevance scoring. Creates up to 5 high-signal suggestions per day. |
| 👤 Manual suggestions | Anyone on the site can suggest a link through a simple form. No repo access needed. |
2. Smart Triage
The moment a suggestion arrives, automation kicks in:
- Solution area detection — AI reads the content and tags it as AIBS, CAIP, Security, or Cross-solution
- Category recommendation — maps the content to one of 10 curated categories
- Reviewer assignment — routes to the right domain expert based on the team roster
- Structured summary — posts a clean summary so the reviewer can decide in seconds
3. Human Review
The assigned expert gets an email with everything they need:
📋 New link suggestion to review
Azure AI Foundry Agent SDK Documentation
🔗 Content: https://learn.microsoft.com/...
📂 Category: Build: APIs & SDKs
🏷️ Solution: CAIP
👤 Reviewer: @you
> Comprehensive SDK documentation covering agent creation,
> tool integration, and multi-agent orchestration patterns.Click the link. Skim it. Comment your decision. That's the entire review workflow.
| Decision | What the reviewer does |
|---|---|
| ✅ Approve | Comments approve |
| ❌ Decline | Comments decline |
| 🔄 Reassign | Mentions the reviewer who should take over |
4. Automated Execution
An approval workflow records the exact URL and issue snapshot, then assigns the Smart Review Agent:
- Validates the URL is reachable
- Checks for duplicates and near-duplicates across the entire library
- Formats the entry with proper metadata (9 structured tags per link)
- Places it in the right category page, grouped logically
- Creates a pull request with a clean commit and immutable approval reference
5. Live in Minutes
After validation passes, the publish workflow verifies the exact approved URL and commit before merging. Azure Static Web Apps deploys, and a notification comments on the original suggestion with the live page URL.
Link suggested ──→ Auto-triage ──→ Expert approves ──→ AI creates PR ──→ Safe publish ──→ Live
(1 min) (instant) (human pace) (2 min) (1 min) (2 min)The Agents
The library runs on a fleet of purpose-built AI agents — some run in GitHub's cloud, others run locally for the curation team.
Cloud Agents (GitHub Copilot Coding Agents)
These run automatically in GitHub when triggered by issues or PRs.
| Agent | What it does |
|---|---|
| 🔍 Smart Review | Executes reviewer decisions — validates URLs, deduplicates, formats entries, creates PRs. The core of the approval pipeline. |
| 🔧 Maintenance Fixer | Reads weekly maintenance reports and creates PRs to fix broken links, update stale dates, and remove duplicates. |
| 🧪 Integration Test | End-to-end pipeline verification — creates test issues and validates the full triage → review → PR → deploy flow. |
| 👤 Partner Review | Reviews the library from a partner persona — evaluates navigation, content usefulness, and discoverability. |
Local Agents (Copilot CLI)
These run in the curation team's terminal for interactive work.
| Agent | What it does |
|---|---|
| 📋 Content Curator | Given a URL, evaluates relevance, checks duplicates, generates a formatted entry, and recommends the right category. |
| 📡 Source Scout | Manages content sources — finds RSS feeds, tests parsing, audits signal quality, generates config entries. |
| 📊 Library Insights | On-demand analytics — coverage gaps, freshness scores, growth trends, quality metrics across all categories. |
Automated Workflows (GitHub Actions)
| Workflow | Schedule | What it does |
|---|---|---|
| Research Agent | Daily, 06:00 UTC | Scans sources, creates link suggestions |
| Auto-Triage | On issue creation | Labels, assigns reviewer, posts summary |
| Link Approval Decision | On reviewer comment | Records the exact approved URL and assigns Copilot |
| Publish Approved Links | After PR validation | Verifies the approval record and merges the exact validated entry |
| Auto-Heal Links | Daily, 06:30 UTC | Repairs safe redirects and raises one issue for uncertain cases |
| Maintenance Agent | Weekly, Mondays 07:00 UTC | Health checks, staleness audit, duplicate detection |
| Link Checker | Daily | Validates all URLs in the library |
| Activity Report | On demand | Produces operational status without routine notification noise |
| Deploy | On merge to main | Builds and deploys the VitePress site |
Category Taxonomy
All content is organized into 10 curated categories, each with a clear scope:
| # | Category | What belongs here |
|---|---|---|
| 1 | Start Here | Role-based entry points and quick-start guides |
| 2 | Agentic AI Fundamentals | Core concepts and shared terminology |
| 3 | Solution Foundations | Foundational content per solution area (AIBS · CAIP · Security) |
| 4 | Architecture & Patterns | Reference architectures, integration patterns, design guidance |
| 5 | Security & Governance | Identity, compliance, governance, responsible AI |
| 6 | Build, Extend & Integrate | APIs, SDKs, tools, tutorials, sample code |
| 7 | Ways of Working | Collaboration, delivery approach, PoC-to-production |
| 8 | Scenario Blueprints | Repeatable end-to-end cross-solution scenarios |
| 9 | Learning & Community | Courses, webinars, recordings, hackathons |
| 10 | News & Updates | Curated high-signal updates only |
Every link carries 9 structured metadata tags: solution area, content type, audience, shareability, maturity, product, pattern, owner, and last-reviewed date. This powers the Browse & Filter page.
Quality Standards
Every link in the library meets these criteria:
- ✅ Relevant — directly useful for agentic AI work across AIBS, CAIP, or Security
- ✅ High quality — official docs, well-written blogs, maintained repos
- ✅ Current — not outdated or deprecated (automated staleness checks)
- ✅ Accessible — publicly available, or clearly tagged as partner-NDA or internal-only
- ✅ Unique — no duplicates (automated deduplication)
- ✅ Well-described — 1–3 sentences explaining what it covers and why it matters
Tech Stack
| Component | Purpose |
|---|---|
| VitePress | Static site generator — fast, searchable, markdown-native |
| Azure Static Web Apps | Hosting with Microsoft Entra ID authentication |
| Azure Application Insights | Anonymous usage analytics (page views, search terms, outbound clicks). Admin-only dashboard. No PII collected, no auth identity sent. |
| GitHub Actions | All automation — triage, research, maintenance, deploy |
| GitHub Copilot Coding Agents | AI agents that operate on issues and PRs |
| Deterministic curation rules | Relevance scoring in the Research Agent without an external model dependency |
Want to Get Involved?
Become a domain expert
We're looking for Partner Solution Architects and Partner Technology Strategists across AIBS, CAIP, and Security to join as domain experts. You'd review link suggestions in your area — typically a few minutes per week.
Contact Johan Wallquist (vTeam Lead) on Teams to get added.
You can also suggest a link right now — no repo access needed.