Our Take · November 20, 2026 · Spencer Schneidenbach

Aviron Labs at Microsoft Ignite 2026

Aviron Labs at Microsoft Ignite 2026

We spent the week at Microsoft Ignite in San Francisco (Moscone). I gave a theater session, hosted a table talk, and spent 8+ hours in conversations with people who are building with AI. That last part is the one I’d actually miss, so I’ll start there.

What people kept asking about

The questions I heard most came back to three things:

  • AI enablement.
  • AI governance: making sure agents are secure, and that users aren’t using tools that have more access than they’re supposed to have.
  • Deploying AI solutions across the broad spectrum of products.

The table talk

I hosted a table talk on what AI agents need before production.

The theater session: RAG, two years later

On Thursday, Nov 19, I gave RAG: Two years later (THR1010): 25 minutes (11:35 to 12:00) in Theater F at Moscone South, Level 200, in the Agent Platform track. Here’s the session page, and the recording is right here if you’d rather watch it.

The full session, from Microsoft. Open the session page if the player doesn’t load.

The short version: the same question, asked two ways, gets very different answers. The demos use a fictional hospital system, Lakeshore Health, with a facilities problem. Everything in them is synthetic. First comes 2024-style vector RAG: retrieve the top-k chunks and answer. Then a 2026 agent that plans, queries, and checks its work.

  • Demo 1, structured data. A C# agent built on Microsoft Agent Framework investigates about 12.5k work orders in Postgres with typed analytics tools, not text-to-SQL. Vector RAG finds tickets that sound like the question. The agent finds the cause: a vendor switch, plus backordered parts, in one building.
  • Demo 2, documents. Azure AI Search agentic retrieval breaks a question into subqueries across contracts, SOPs and meeting minutes. The agent combines that with the work-order evidence and drafts an escalation with citations. The same agent also runs as a Microsoft Foundry hosted agent.

That’s the whole gap in one sentence, tbh. One approach finds text that resembles your question. The other goes and checks.

Run it yourself

The code is public: schneidenbach/rag-two-years-later.

  • Replay only needs no Azure account, just .NET 10 and Node.
  • The full live setup creates billable Azure resources (a General Purpose Postgres server and a Basic Azure AI Search service), so delete the resource group when you’re done.
  • The repo includes a Claude Code skill that handles setup: open it in Claude Code and run /setup-lakeshore.

Working through any of this?

If you’re working through AI enablement, governance, or getting agents to production, get your AI game plan. You’ll talk to the people who’d actually do the work. If you want to start with the codebase, that’s what our AI Agent Readiness review looks at: tests, CI, docs and conventions, environment setup, secrets, evals.

S
Spencer Schneidenbach

Spencer Schneidenbach is the CTO of Aviron Labs. He speaks on production AI, agentic coding, evals and AI security.

More about Spencer

Get your AI game plan.

A free hour with our problem solvers.

Get your AI game plan