- Subject
- Microsoft 365 & SharePoint
- Filed
- Source
- Koskila
- Loans
- 791
aMP Rennes 2026 Session to Build Microsoft AI Agents with Foundry and Copilot
A 45-minute session in Cesson-Sévigné will examine Microsoft Foundry, M365 Copilot and agent design through live coding, demos and architecture trade-offs.

Microsoft AI agents need more than a prompt and a connection to a language model. At aMP Rennes 2026, a live build will examine how Microsoft Foundry, Microsoft 365 Copilot, Copilot Studio and the Microsoft 365 Agents SDK fit together when the goal is to support real work.

The session, “M365 Copilot is the UI, Foundry is the Brain, but what turns intelligence into action?”, is scheduled for Wednesday, October 28, from 2:00 to 2:45 p.m. in Room 2 at Orange Atalante, 4 rue du Clos Courtel, 35510 Cesson-Sévigné. The presenters are the author and Michaël Maillot.
Attendees can expect live coding, multiple demos and practical lessons from building systems. The presenters also promise a few questionable decisions made for learning purposes, so the session is designed to show the engineering choices and trade-offs behind an agent, not just its finished interface.
Where each part of the Microsoft AI stack fits
The session’s framing gives M365 Copilot the role of user interface, while Microsoft Foundry provides the environment for building and coordinating the underlying intelligence. The Microsoft 365 Agents SDK connects those systems back into Microsoft 365, Teams and Copilot. Copilot Studio is part of the discussion too, particularly when deciding whether a low-code approach can meet the need or whether a code-first design is justified.
That division matters for teams evaluating AI automation. A conversational interface may make an agent easy to reach, but the work behind it can involve model selection, evaluation, tools, orchestration and memory. The presenters plan to build an agent from the ground up and use it to examine those components in context.
The session will also look at Work IQ, custom engine agents and agent-to-agent collaboration. These topics point beyond simple question-answering: the intended focus is on systems that can draw on relevant context, reason through a task and take useful action in operational workflows.
Engineering trade-offs, not just demos
Manufacturing and industrial operations are among the settings highlighted for investment in AI-powered automation, operational intelligence and decision support. In those environments, a useful agent has to do more than produce a plausible response. It must fit a process, use appropriate context and help people make informed decisions.
Model choice is one design question. The presenters will discuss when different models, evaluations, orchestration patterns or memory can improve an outcome, and when those additions create complexity without enough benefit. For administrators and developers, that distinction affects maintainability as well as the work required to build and operate an agent.
Another decision is where to draw the line between Copilot Studio and custom code. Low-code tooling can be a productive fit for some needs; the session will consider the point at which its limits make a code-first approach more useful. The emphasis is on recognizing that boundary before an architecture becomes difficult to maintain.
Architecture that works in a short demonstration may not be the right choice for a system that has to scale or support ongoing business processes. The planned discussion covers patterns that scale, patterns that remain maintainable and approaches that look impressive in a demo but may not hold up in regular use. That makes the live build useful as a way to expose trade-offs, rather than treating a successful demonstration as proof that a design is ready for operational use.
Session details and takeaways
The event is the author’s first aMP event and first visit to France. The author also describes it as a first opportunity to speak at an event there. The personal setting aside, the practical value for attendees is a focused look at how Microsoft’s agent tools can be combined and where additional engineering effort makes sense.
By the end of the 45-minute session, attendees are expected to have a clearer sense of where Foundry can add value, when Copilot Studio is sufficient, and how to avoid costly architectural choices. They’ll also see how Work IQ, custom engine agents and Foundry can contribute to automation and operational intelligence when the goal is an outcome rather than another conversational interface.
The event agenda and details are available on the aMP Day Rennes 2026 website. For teams assessing Microsoft’s agent stack, the session offers a chance to compare the tools through code, demonstrations and the practical decisions that shape a maintainable system.