It’s so cool how the brain works. You can be staring at a problem in 2026, and suddenly you remember something you read way, way back that kind of saves your ass. For me, that moment happened last...

It started with a simple piece of curiosity: when I ask GitHub Copilot to create a dynamic workflow, how does it actually work behind the scenes? You type a plain-English sentence — “build me a wo...

I spent the weekend digging through LangGraph’s source code, trying to understand what actually happens beneath the abstractions. These are my notes, the experiments I ran, and the mental model I e...

Five practical examples of when durable execution makes sense, when it doesn’t, and what it means for AI agents. After my last couple of posts about Durable Task, I got a comment that made me real...

In the previous post, we followed an orchestration through its activity awaits and looked at what actually survives: history, not the original worker’s local tasks or suspended method. I ended by p...

I’ve recently joined the Durable Task team under Azure Serverless, and I’m very excited. Durable tasks are such an interesting mix of complexity and elegance. I’ve written about Durable Task and D...

This is the real-world version of the pattern I wrote about in When the Cluster Stops Owning the Inner Loop, and Why Aspire Is Quietly Disrupting Platform Engineering: a Kind cluster (Kubernetes...

This post builds the smallest version of the loop that still shows the whole thing: a real Kind cluster, a real CRD, and a Go reconciler running where I can debug it. In follow-up posts I take t...

The most common question I get about Squad is some version of the same one: can you actually show me — with data — that coordinating a team of agents adds real value over just pointing one good mod...

In Part 1, I put Squad inside a deterministic LangGraph workflow. In Part 2, I moved risky tool execution into Azure Container Apps Dynamic Sessions. That gave me a clean rule: the app owns the wo...