A useful Azure reference architecture is less about listing services and more about making trust boundaries, entry points, and data paths obvious.
Traffic charts become useful when they help engineering, support, and product decide what to investigate, fix, or explain next.
What iterative AI image prompting taught me about generating usable enterprise visuals instead of generic AI art.
A delivery process does not improve because the diagram looks cleaner. It improves when ownership, handoffs, and the definition of done become harder to misread.
How I built a TypeScript CLI for Microsoft To Do in a single session with Copilot, packaged it with GitHub Actions, and integrated it as a shared OpenClaw skill so every agent can manage my task lists.
How I expanded my OpenClaw gateway from two agents to five, routed each to its own Teams channel, and discovered why RSC permissions matter more than config files.
Microsoft Agent Framework 1.0 just went GA. Here's what actually matters in it: stable APIs, A2A, declarative definitions, and what it still doesn't answer for you.
A small routing detail broke OpenClaw reminder delivery in exactly the wrong place. Here's how I traced it, fixed same-channel delivery, and stopped the agent from claiming success when scheduling had actually failed.
How I split OpenClaw into main and personal agents, gave them separate workspaces, and routed Signal and Teams differently so the system felt more natural to live with.
Why complex enabling work often disappears in large organizations, and why I now think of visibility as part of delivery rather than as a political side quest.
