Samuel Tambunan's Blog

Weekly .NET & AI Briefing — September 12–19, 2026

This week’s roundup leans into two threads that matter for the architect/leadership track: performance and resilience engineering at scale, and the fast-moving overlap between AI agents and everyday application code. On the .NET side, Microsoft shipped a deep performance retrospective for .NET 11 and a live series on building AI agents in C#, while Andrew Lock and Milan Jovanović both published sharp, practical pieces on security and data-layer design decisions. On the big-tech side, Uber had a strong week for architecture write-ups (retry storms, service-mesh dependency analysis, search pipeline rebuilds), Elastic shipped notable AI-tooling and storage-engine posts, and both Netflix and Airbnb published on platform and AI-infrastructure engineering. A few of the specific names you asked about — Hanselman, Fowler, Gravell, Newton-King — didn’t have new long-form posts land in this exact window, so they’re not represented this week; worth checking back next time.

.NET & C# Engineering

  1. Build Your Own AI Agent Harness in C#, the MAF Claw Live Series — .NET Blog (Bruno Capuano), September 16 Bruno Capuano is building a complete AI agent harness live in C#, starting from a single call around IChatClient and working up to a production-ready, observable, governed agent. Good primer if you want hands-on grounding in how agent orchestration actually gets built rather than just consumed via a framework.

  2. Performance Improvements in .NET 11 — .NET Blog (Stephen Toub), September 15 Stephen Toub’s annual deep-dive catalogs hundreds of runtime, JIT, and library-level performance improvements landing in .NET 11. It’s long, but as an architect you want the mental model of why things got faster (allocation reduction, vectorization, tiered compilation changes) more than the changelog itself.

  3. Use C# Unions and Closed Hierarchies in ASP.NET Core — .NET Blog (Dmitrii Korolev), September 10 A practical look at using the new C# union/closed-hierarchy features for modeling API results and discriminated responses in ASP.NET Core. Relevant if you’re designing typed, exhaustive result models for service boundaries.

  4. Understanding Device Bound Session Credentials (DBSC) — Andrew Lock, .NET Escapades, September 15 Andrew Lock explains DBSC, a browser/session security mechanism designed to make stolen session cookies useless by cryptographically binding sessions to a device. Useful architecture-level context if you own auth/session design and want to understand where the industry is heading beyond bearer tokens.

  5. Let Postgres Enforce Tenant Isolation — Milan Jovanović, September 12 A concrete walkthrough of using PostgreSQL Row-Level Security with EF Core to push multi-tenant data isolation down into the database rather than relying purely on application-layer filtering. A good pattern to have in your back pocket for multi-tenant SaaS architecture reviews.

  6. Adding an Index Made This Query Slower — Milan Jovanović, September 19 A short, memorable case study on how adding an index can cause Postgres’s query planner to pick a worse execution plan — in this case scanning half a million rows to return ten. A useful reminder for anyone doing database performance reviews that “more indexes” isn’t always the fix.

Big Tech Engineering & AI

  1. How Uber Protects Against Retry Storms — Uber Engineering, September 17 A resilience-engineering write-up on how Uber prevents cascading retry amplification from taking down downstream services during partial outages — core reading for anyone designing fault-tolerant distributed systems.

  2. Halving the Time: How Uber Eats Rebuilt Its Search Pipeline — Uber Engineering, September 10 A case study on re-architecting a high-traffic search pipeline for latency, covering the trade-offs made to cut response time roughly in half. Good material for thinking about search/retrieval architecture at scale.

  3. Large-Scale Automated Dependency Analysis Across Uber’s Service Mesh — Uber Engineering, September 15 Covers the tooling Uber built to automatically map and analyze service dependencies across its mesh — directly relevant if you’re thinking about how to maintain architectural visibility as a system grows past the point any one person can hold it in their head.

  4. You and Your AI Agent Shouldn’t Be Using curl: Introducing the Elastic CLI and Agent Skills — Elastic Search Labs (Josh Mock), September 14 Elastic’s take on giving AI agents a purpose-built CLI and skill system instead of letting them shell out to raw HTTP calls — a good concrete example of the “tool design for agents” problem that’s becoming a real architectural concern.

  5. Columnar Storage Isn’t a Columnar Database: What Columnar Mode Brings to Elasticsearch — Elastic Search Labs (Yannis Roussos), September 15 A storage-engine deep dive distinguishing columnar storage layout from being a full columnar database, using Elasticsearch’s new Columnar mode as the example. Solid for sharpening your mental model of storage engine trade-offs, which transfers directly to data-layer architecture decisions.

  6. Leave the Class Path in the Rearview Mirror — Netflix Technology Blog, September 18 Netflix on moving away from the classic JVM classpath model toward the module path, and the platform-engineering work required to migrate a large, long-lived codebase. Even outside the JVM world, it’s a good read on the mechanics and organizational discipline of large-scale platform migrations.

  7. Beyond the Model: Engineering AI Infra with Scientific Judgement — Airbnb Tech Blog, September 2026 Airbnb’s engineers argue that building reliable AI infrastructure takes as much “scientific judgement” — knowing when to trust a model, when to add guardrails, how to design evaluation — as it does raw engineering. A good leadership-lens read on how to structure teams and decisions around AI features rather than just the AI itself.