Articles & Deep-dives

Microsoft Scout signals a shift from Copilot-as-assistant to Copilot-as-colleague — an agent that works ambiently in the background instead of waiting for a prompt. For architects, that raises real questions before rollout: what triggers the agent, what data scope it's allowed to read, and how you audit what it acted on. This piece walks through what changes for enterprise governance once AI starts acting without being asked.

The Configurable Form Engine already lets admins define Dynamics 365 form behavior — field visibility, validation, defaults — entirely through Dataverse config records, no JavaScript required. This follow-up adds a Copilot layer on top: describing the form change you want in plain English and having it translate into the same config records the engine already understands. It's a look at what "low-code" actually means once an AI agent is the one writing the configuration.

Most "AI in your CRM" pitches are still just a chatbot bolted onto a knowledge base — you ask, it answers. This article makes the case for event-triggered agents instead: automation that reacts to what's actually happening in Dynamics 365 — a stage change, a stale opportunity, a missed SLA — and takes multi-step action on its own. That's the practical difference between AI as a search box and AI as a coworker.

Every Dynamics 365 project reinvents the same form JavaScript — show this field, hide that section, default this value — project after project, client after client. This article breaks down the Configurable Form Engine's approach: moving that logic out of code and into versioned Dataverse configuration, so a form behavior change is a data update, not a deployment.

Most AI coding assistants have one instinct: you ask, they build. This article explains why the Power Platform Architect Skill for Claude Code is built to do the opposite first — pushing back with clarifying questions about schema, security roles, and solution layering before generating anything, the way a senior architect would. It's a case for slowing an AI agent down being a feature, not a bug, in Power Platform delivery.

Microsoft 365 Copilot is only as useful as the context it can reach. Model Context Protocol (MCP) is Microsoft's answer to that limit — a standard way to wire Copilot into systems it wasn't natively built to see, from internal databases to line-of-business apps. This article breaks down what MCP actually does and where it fits alongside Copilot's existing connectors.

Large language models are only useful against enterprise data if they can query it safely and correctly. This preview walks through connecting Dataverse to an LLM via MCP — what the server needs to expose, how authentication is handled, and where the guardrails belong so a language model can query CRM data without becoming a security hole.

Managed Environments turn on a set of governance controls most Power Platform admins don't fully use — usage insights, sharing limits, weekly digest emails, and pipeline-based deployment. This guide walks through what actually changes when you enable them, and the lifecycle-management decisions that determine whether they help or just add friction.

MCP is quickly becoming the common language between AI models and the tools they need to act on — the same role REST APIs played for web services. This article explains why Microsoft is standardizing around it, what a client-server MCP integration actually looks like, and why that matters for anyone building AI agents on the Microsoft stack rather than just chat interfaces.

The Copilot experience inside Model-Driven Apps is still evolving, and it's not always clear where the customization boundaries actually sit today. This deep-dive covers what you can configure — and what's still locked down — across the current Model-Driven App Copilot experience, so you know what to promise a client before you scope the work.