73 posts, newest first.
The model is the brain, the harness is everything else: instructions, tools, memory, permissions, approvals, logs and stop rules. Why the harness, not the model, runs the tool calls and decides whether an agent works.
How I built a personal Second Brain in OneDrive with PARA and Copilot Cowork that handles the hardest part: routing every note to the right place automatically.
Most agent projects don't fail because of the tech. They fail because of unrealistic expectations, bad data, and wrong platform choices.
No governance doesn't mean chaos — it means IT blocks everything. My framework for the six pillars of agent governance.
Everyone is building agents, but almost nobody is testing them properly. My take on new testing categories, tooling, and why ALM for agents is overdue.
The next step isn't better agents — it's agents as teammates. My 5-stage maturity model for the journey to an agentic organization.
When does extending Microsoft 365 Copilot actually pay off? My take on real ROI — beyond time savings and Hello World demos.
Microsoft's promise that business users can independently build production-ready Copilot agents through Agent Builder and Copilot Studio breaks down the moment you need real integrations, complex debugging, or enterprise governance. Instead of selling the illusion of fully self-service agent development, organizations should embrace a tiered model where citizen developers prototype ideas that are refined collaboratively with technical support.
Microsoft 365 Copilot can be extended through three pillars—Connectors (for bringing external data in), Agents (for performing actions via OpenAPI, MCP, or Copilot Studio), and APIs (for embedding Copilot capabilities in your apps)—but the key gotcha is that Actions only work inside Agents. Start with Declarative Agents for quick wins, use Graph Connectors for read-only data access, and remember that proper data governance must be in place before connecting any data sources
The workplace is evolving from prompt-based assistants to an agent-centric model, where specialized digital teammates pursue goals autonomously. Microsoft’s ecosystem—from SharePoint to Azure AI Foundry—empowers everyone to build agents that seamlessly extend Microsoft 365 Copilot. What began with the Bot Framework’s Skills is now multi-agent collaboration across platforms. The future isn’t Copilot in apps but Copilot across work—agents like a Project Planner that follow you everywhere. As we shift from Copilot adoption to agent lifecycle management, one question remains: What’s your vision for the agentic workplace?
Examining how Microsoft Copilot Studio's preview release of autonomous agent capabilities lays a basic yet promising foundation—highlighting its core architecture, current functionalities, and future potential for automating enterprise workflows.
Enabling deep reasoning on a Microsoft Copilot Studio agent and use it in Microsoft 365 Copilot
