In this episode of M365.fm, Mirko Peters explains why hand‑rolled AI agents in Microsoft 365 always look fine in a demo and then fall apart in production — and why the Microsoft 365 Agent SDK is now the minimum architecture, not a “nice to have.”

WHAT YOU WILL LEARN

• Why DIY agents break on the basics: identity, state, channels, governance, and debugging
• How “app‑only everywhere” destroys permission fidelity, audit trails, and user trust
• Why stateless bots forget context as soon as you add load balancers, multiple nodes, or tool calls
• How channel differences (Teams, web, Slack, Outlook, Copilot Studio) quietly wreck UX if you reinvent adapters yourself
• How the Agent SDK standardizes identity, state, protocol, and delivery so you can focus on cognition and tools
• What the SDK actually gives you: proper auth, durable conversation state, multi‑channel adapters, streaming, diagnostics, and orchestrator neutrality
• A step‑by‑step implementation blueprint: from “hello world” to a multi‑channel, tool‑using agent that passes security review
• How Purview, DLP, Defender, and Zero‑Trust controls plug into agents when you build on the SDK instead of raw webhooks

THE CORE INSIGHT

Most M365 agent projects fail for boring reasons, not AI reasons. They die on sign‑in flows, lost state, broken Teams cards, untraceable errors, and “Who approved this permission?” questions — long before model quality is even discussed. The Microsoft 365 Agent SDK is the missing foundation: it handles identity, state, channels, and governance so your “agent” behaves like a first‑class citizen of your tenant instead of a side project with production access.

WHO THIS EPISODE IS FOR

This episode is ideal for Microsoft 365 architects, platform engineers, AI teams, and security/governance leads who are under pressure to ship agents into Teams, Copilot Studio, and web chat without creating a parallel, fragile shadow platform. If your current agent prototypes are a tangle of web APIs, custom adapters, and undocumented permissions, this conversation gives you a concrete path to rebuild on the Agent SDK before you scale.

ABOUT THE HOST

Mirko Peters is a Microsoft 365 consultant and digital workplace architect focused on building safe, governable AI agents and automation on the Microsoft cloud. Through M365.fm, Mirko shares practical architectures, SDK patterns, and real‑world lessons that help teams move from lab‑grade bots to production‑ready agents that security and operations can live with.

Become a supporter of this podcast: https://www.spreaker.com/podcast/m365-fm-a-microsoft-mvp-podcast-by-mirko-peters--6704921/support.