Trascritto
Copilot Custom Agents: Copilot Is Broken Until You Do This
22 nov 2025 ·
21 min. 31 sec.
Scarica e ascolta ovunque
Scarica i tuoi episodi preferiti e goditi l'ascolto, ovunque tu sia! Iscriviti o accedi ora per ascoltare offline.
Copilot Custom Agents: Copilot Is Broken Until You Do This
Capitoli
Descrizione
Copilot sounds confident but falls short when a question depends on your own DLP exceptions, regional SOPs or compliance memos. This episode shows Microsoft 365 architects and developers how to...
mostra di più
Copilot sounds confident but falls short when a question depends on your own DLP exceptions, regional SOPs or compliance memos. This episode shows Microsoft 365 architects and developers how to build a custom engine agent and bring it into Copilot Chat with a small manifest change. You get a build, deploy and verify plan for faster, cited and more accurate answers.
In this episode
Key takeaways
Worth quoting
“Default Copilot doesn't know your exception list. It doesn't know your project Orion carve out. It can describe DLP in theory, it can't apply your DLP in practice. That gap is where incidents happen.” – Mirko Peters (01:17)
“Because out of the box, Copilot is optimized for breadth, not your depth. It's grounded by Microsoft Graph and Public Knowledge, which is the right default for most people, but the wrong default for enterprise precision.” – Mirko Peters (02:26)
Explore more on M365.FM
Follow and rate M365.FM in your podcast app, and join the free M365 community at m365con.net.
Become a supporter of this podcast: https://www.spreaker.com/podcast/m365-fm-microsoft-365-azure-cloud-power-platform-ai--6704921/support.
mostra meno
In this episode
- 00:00 – Copilot the confident generalist
- 00:33 – Where default Copilot fails
- 04:21 – Custom engine agents as specialists
- 08:04 – The two-minute manifest tweak
- 11:05 – Build, deploy, verify
- 16:39 – Before versus after: three proof cases
- 20:17 – Life cycle, governance and scale
Key takeaways
- Default Copilot is grounded in Microsoft Graph and public knowledge and cannot apply company-specific DLP rules, SOPs or compliance memos on its own.
- A custom engine agent adds retrieval with Azure AI Search, actions against internal APIs and guardrails such as tenant controls and data scopes.
- Updating the app manifest to schema 1.22 with the right Copilot agent capabilities brings the specialist into Copilot Chat, where it answers with citations.
Worth quoting
“Default Copilot doesn't know your exception list. It doesn't know your project Orion carve out. It can describe DLP in theory, it can't apply your DLP in practice. That gap is where incidents happen.” – Mirko Peters (01:17)
“Because out of the box, Copilot is optimized for breadth, not your depth. It's grounded by Microsoft Graph and Public Knowledge, which is the right default for most people, but the wrong default for enterprise precision.” – Mirko Peters (02:26)
Explore more on M365.FM
- Episode page: the full episode with chapters, quotes and FAQ on custom engine agents
- Microsoft 365 Copilot – the confident generalist that cannot apply company-specific DLP rules, SOPs or memos
- Microsoft 365 Copilot Chat – where the custom engine agent appears after the manifest update, inside the familiar UI
- Data loss prevention (DLP) – the first proof case, checking external sharing against a tenant's exception list
- Related episode: Stop Typing to Copilot: Use Your Voice NOW!
Follow and rate M365.FM in your podcast app, and join the free M365 community at m365con.net.
Become a supporter of this podcast: https://www.spreaker.com/podcast/m365-fm-microsoft-365-azure-cloud-power-platform-ai--6704921/support.
Informazioni
| Autore | Mirko Peters M365, Azure & AI |
| Organizzazione | m365 FM |
| Sito | www.m365.fm |
| Tag |
Copyright 2026 - Spreaker Inc. an iHeartMedia Company
Commenti