Declarative Agents vs. Custom Engine Agents: One Needs a Copilot License, One Doesn’t — and Copilot Studio Builds Both

Declarative agents run on Microsoft's orchestrator and require a Microsoft 365 Copilot license for every user — but cost nothing to host. Custom engine agents require you to supply the orchestration layer and pay for hosting — but your users need no Copilot license at all. This post explains the architectural split behind that licensing inversion, shows how to read either agent type from the app manifest, resolves the Copilot Studio taxonomy confusion (Agent Builder always produces declarative agents; the full platform can produce either), and provides a verified decision framework — including three binary differentiators and an eight-question pre-commit checklist — sourced from the official Microsoft 365 extensibility documentation updated August 2026.

Microsoft 365 Copilot Connectors: Two Connector Types, Three Mandatory Labels, and the One Admin Step Most Deployments Miss

Microsoft 365 Copilot connectors come in two architecturally distinct types — synced connectors that index a copy of external content into the Microsoft 365 semantic index, and federated connectors that fetch data live via MCP at query time — and custom connectors can only be built for the synced type. This post covers the four-step custom connector build lifecycle, the three mandatory semantic labels (iconUrl, title, url) that gate Copilot surfacing, the admin inline-results step that most deployments miss, the licensing table that determines which experiences your tenant can actually use, and a seven-point pre-go-live checklist that catches silent failures before they reach production.

Subscribing to Copilot Activity: The AI Interactions Change Notifications API, Dual Subscription Scopes, and What Best-Effort Delivery Means for Your Audit Pipeline

The AI Interactions Change Notifications API delivers per-interaction Copilot events via the standard Microsoft Graph webhook pattern — but the dual subscription scope model, a hard `lifecycleNotificationUrl` requirement that fails subscription creation if absent, and a best-effort delivery model with no recovery path after a four-hour retry window mean a generic webhook implementation will break in ways that are not obvious until they cost you data. This post covers every required field, both notification payload shapes, the throttling thresholds that drop events without warning, and a nine-point go-live checklist that catches silent failures before your audit pipeline goes live.

Microsoft 365 Copilot Meeting AI Insights API: What callAiInsight Actually Returns and How to Retrieve It

The Microsoft 365 Copilot Meeting AI Insights API exposes `callAiInsight` — a resource that went generally available in December 2025 and delivers three data types (AI-generated meeting notes, action items with a named owner, and caller-specific mention events) through a mandatory two-call retrieval pattern that most documentation and community content does not describe correctly. The list call returns only five metadata fields; the get call returns the actual content — and the integration gaps this creates, from the `ownerDisplayName` plain-string issue to the Global-service-only constraint, are what make this API non-trivial to ship.

Microsoft 365 Copilot Chat API: Work IQ (GA) vs Graph Beta — Which One to Build Against

As of August 2026, calling Microsoft 365 Copilot programmatically means choosing between two distinct implementations — the Work IQ Chat API (general availability, one permission, Copilot Credits billing) and the Graph beta Chat API (preview, seven required delegated scopes, flat M365 Copilot license) — and that choice shapes your authentication architecture, your enterprise admin consent conversation, and your billing model before you write a line of code. This post covers the complete request and response shape for both paths, the permission and billing models that cannot be swapped at go-live, and the five things neither implementation can do regardless of which path you choose.

The Microsoft 365 Copilot Retrieval API: What Every Developer Must Know Before Building a RAG Integration

The Microsoft 365 Copilot Retrieval API gives developers a direct path to retrieval-augmented generation on organizational content — no chunking pipeline, no vector store, no sync job — by exposing the semantic index Microsoft 365 already maintains for Copilot across SharePoint, OneDrive, and connector content. This post covers the complete request and response shape, the delegated-only auth constraint that rules out daemon integrations, the 200-request-per-user-per-hour rate limit strategy, and a clear decision boundary between the Retrieval API and the beta Search API.