Copilot Custom Agents: Copilot Is Broken Until You Do This

(00:00:00) The Limitations of Default Copilot
(00:00:32) The Need for Custom Engine Agents
(00:04:40) The Three Pillars of Authority
(00:05:01) Building a Custom Engine Agent
(00:07:33) Implementing the Specialist in Copilot Chat
(00:09:39) Verification and Testing
(00:19:11) Quantifying the Improvement
(00:20:11) Scaling and Governance

n this episode of M365.fm, Mirko Peters explains why out‑of‑the‑box Microsoft 365 Copilot fails on real‑world enterprise questions — and how custom agents turn it from a clever generalist into a governed specialist that actually follows your rules.

WHAT YOU WILL LEARN

• Why default Copilot gives “nice” but wrong answers about your policies, DLP exceptions, escalation paths, and regulated processes
• How Copilot’s standard grounding (Graph + public info) misses local reality: your playbooks, exceptions, SLAs, and approval rules
• What custom engine agents are: specialized brains connected to your own indexed content, APIs, and tools
• How a custom agent uses retrieval (Azure AI Search), tools (internal APIs like CheckOnCallSchedule or ValidateCustomerId), and guardrails to answer correctly
• Why upgrading your manifest to schema 1.22 and adding copilotAgents/customEngineAgents is the key step most tenants are missing
• How to design narrow, high‑value agents (for support policy, HR, security, or operations) instead of one “do everything” monster
• How to run agents as products: environments, versioning, evaluation, and clear ownership

THE CORE INSIGHT

Copilot isn’t broken — it’s blind to your world. By default, it doesn’t know your exception lists, approval chains, escalation rules, regional variants, or internal APIs, so it answers from generic Microsoft patterns and best practices. That works for low‑risk questions and fails spectacularly when users ask, “Are we allowed to…?” or “What is the process here?”Custom agents fix this by giving Copilot a specialist to talk to. Instead of guessing, Copilot routes the hard questions to an agent that can search your curated content, call your systems through safe tools, and then return grounded, policy‑correct answers with clear citations. The moment you upgrade your manifest and wire in a custom engine agent, Copilot stops improvising on critical topics and starts behaving like part of your operating model.

WHO THIS EPISODE IS FOR

This episode is ideal for Copilot program owners, Microsoft 365 architects, platform engineers, and governance or compliance leads who are responsible for making Copilot safe and useful in the enterprise. If your users love Copilot’s potential but you don’t trust its answers on policy, security, or process, this conversation gives you a clear blueprint for implementing custom agents the right way.

ABOUT THE HOST

Mirko Peters is a Microsoft 365 consultant and digital workplace architect focused on building governed, agent‑ready environments on the Microsoft cloud. Through M365.fm, Mirko shares practical architectures, manifest patterns, and real‑world stories that help organizations turn Copilot from a clever demo into a reliable, policy‑aware assistant.
























Become a supporter of this podcast: https://www.spreaker.com/podcast/m365-fm-modern-work-security-and-productivity-with-microsoft-365--6704921/support.