DISPATCHES FROM THE MICROSOFT AI ECOSYSTEM
This post walks through a video session by enterprise architect Hallowell M. Longley, "Agentic AI Architecture" (youtu.be/IcEqtjrhyB0). It covers Microsoft Foundry, Zero Trust, and SASE, and how they fit together when you run AI agents at scale. The main idea is simple. The old security perimeter is mostly gone. Agents now run across many clouds, so you need one place to build and manage them, and one set of security rules that follows them. This post does not cover pricing, licensing, or setup steps. I checked the session's Foundry claims against Microsoft Learn
RAG has always had a dirty secret: it only searches one index at a time, and real questions never stay inside one index. Foundry IQ tries to fix that with a managed knowledge layer that plans, decomposes, and merges across sources. This is a reference guide to what it actually does, what it costs on two separate bills, and where the limits are still moving under you.
An AI agent isn't just a chatbot with more steps. The real difference is a control layer wrapped around the model. That layer decides what tools the agent can use, how much it remembers, when it needs to ask permission, and whether anyone can trace what it actually did. This layer is called the agent harness.
This final article puts the rewrite to a real test: production-scale volume, an actual Task Reviewer pass that catches something the earlier work missed, and a benchmark that settles whether the win holds up once the row count moves from a few hundred to over ten thousand.
This article puts Hyper-Velocity Engineering to work on a complex rewrite: turning the RBAR cursor we described in the previous article, "Temporal Tables in a Real Database (Not a Toy Example)", into a single set-based query, without breaking the loader downstream.
WideWorldImporters ships with temporal tables that version themselves for free, no ETL code required. But there's a catch: the built-in syntax can tell you what a row looked like at any moment, and nothing about what changed since your last load. This article digs into that gap, and the two pieces of hand-written logic WideWorldImporters uses to bridge it.
Everyone's demoing Hypervelocity Engineering with 3x velocity numbers and a "30% of Microsoft's code is agent-written" line nobody can source. Here's what's actually underneath it: a four-agent workflow that won't let AI write code until it's done its homework, plus the install method most writeups get wrong (hint: it's not just VS Code).