Notes from real projects. What worked, what didn't, and why — no marketing spin.
Microsoft is moving Claude processing in Microsoft Copilot into the EU Data Boundary by the end of 2026. Why that removes the strongest blocker for European tenants, why the decision still belongs to you, and how architects, DPOs, admins, and works councils should run it.
Copilot Studio now has three harnesses, and the new Customer Conversation harness is built for voice and chat with customers. What it means for architecture, memory, biometrics, escalation, cost, and governance.
A living article on Copilot and agent costs. Where Copilot, Cowork, Copilot Studio, and Foundry agents show their costs, which controls actually stop spending, how to charge costs back, and what Microsoft has changed recently.
A living article on what Copilot costs. What the Copilot license costs, how Copilot Credits work, which features use them, and what Microsoft has announced next.
A practitioner's reference for Power Platform core, Copilot Studio, Dataverse, Dynamics 365 Contact Center, and Dynamics 365 Customer Service.
With agents, a single request can fan out into model calls, retrievals, retries, and tool actions that cost far more than the prompt suggests. Here's how I measure, allocate, and govern that full cost — not just the token line — and tie it to outcomes.
This write-up shows how to build Skills in the new Copilot Studio agent experience. It covers the SKILL.md format, bundling scripts and resources, authoring and uploading, how the orchestrator uses them, and governance. It is separate from the legacy Bot Framework skill.
A deep dive into the rebuilt Copilot Studio — its natural-language authoring surface, the new agentic orchestrator, Memory, Microsoft IQ, and the rise of SKILL.md as an open, behavior-as-code pattern for enterprise agents.
Microsoft rebuilt Copilot Studio at Build 2026: a four-surface agent model, Skills as Markdown, a Memory surface, a new agentic orchestrator, a new workflow designer, and convergence with Microsoft Foundry. What changed, what it means for the patterns in Parts 1 and 2, and how to think about the move.
Four end-to-end scenarios on different orchestration patterns, a starter MCP server inventory, and the ten anti-patterns that recur in production Copilot Studio deployments.
A long-form, end-to-end foundation for building multi-agent systems in Microsoft Copilot Studio — when to split, the five surfaces, four patterns, and the two open standards (MCP and A2A) that make the whole thing composable.
I connect enterprise AI architecture with practical implementation, and share what I learn along the way.
I’m a Principal Architect at NTT DATA, Microsoft MVP and Microsoft Certified Trainer (MCT), based in Berlin. My work focuses on agentic AI, intelligent automation, and the Microsoft ecosystem. I help organizations move from promising demonstrations to secure, scalable, well-governed solutions. That means connecting technology choices with business processes, enterprise data, operational ownership, and measurable outcomes.
