Samuel Tambunan's Blog

Weekly .NET & AI Briefing — September 19–26, 2026

This week’s roundup leans hard into the theme running through almost every post below: agentic AI is no longer a side experiment, it’s showing up in ASP.NET Core protocols, Elasticsearch operations, and the daily workflow of senior engineers. Alongside that, there’s solid platform news out of .NET 11 and a handful of “how we actually run this at scale” posts from Netflix, Uber, Elastic, and Airbnb that are worth studying for architecture patterns. Fourteen links, roughly 90 minutes of reading, ordered so you can skim the .NET platform news first and save the systems-design posts for when you have a quiet block of time.

.NET 11 & Platform Updates

  1. Performance Improvements in .NET 11 — .NET Blog, September 15 The annual deep-dive into the runtime, JIT, and library work behind .NET 11’s performance gains. Worth reading closely if you’re making the case internally for upgrading — it’s the kind of detail that turns “let’s upgrade” into a defensible architectural decision.

  2. Experimental support for Device Bound Session Credentials (DBSC) in ASP.NET Core — Andrew Lock, September 22 Andrew Lock’s ongoing ”.NET 11 preview” series turns to DBSC, a browser mechanism for binding session cookies to a device to blunt cookie-theft attacks. Good context for anyone thinking about auth architecture beyond just “add another JWT.”

  3. Creating a memory dump in C# — .NET Blog, September 22 A practical, tools-focused walkthrough of capturing process state for post-mortem debugging — the kind of production-diagnostics skill that separates senior engineers from architects who can actually support what they design.

AI Meets .NET

  1. AG-UI Protocol now has a first-class .NET SDK — .NET Blog, September 25 Microsoft and CopilotKit shipped an MIT-licensed .NET SDK for AG-UI, the emerging protocol for streaming typed events between agents and user-facing apps. If you’re going to architect agentic features into a .NET product, this is the interoperability layer to know.

  2. Build Your Own AI Agent Harness in C#: the MAF Claw Live Series — .NET Blog, September 16 A four-part series on building a production-ready agent harness with real observability, not just a demo loop. Useful reference architecture if leadership starts asking “can we build this ourselves instead of buying it.”

  3. I Replaced a $120/Year SaaS With My Own App — Milan Jovanović, September 26 A small, honest case study: Milan uses a coding agent plus AssemblyAI’s API to replace a paid transcription tool with a self-built .NET app in under an hour, dropping annual cost from $120 to under $8. A nice, concrete data point on where AI-assisted development actually pays off.

Practitioner Deep Dive

  1. Adding an Index Made This Query Slower — Milan Jovanović, September 19 A sharp explainer on Postgres’s query planner picking a worse index and why “just add an index” is bad architectural advice without understanding selectivity and cost estimation. Directly useful the next time a team lead proposes an index as a silver bullet.

Big Tech Engineering Blogs

  1. Netflix Stratum Media Processing: Automated Container Right-Sizing — Netflix TechBlog, September 25 How Netflix automatically right-sizes containers for its media-processing pipeline instead of relying on static resource requests — a good pattern to borrow for any team drowning in over-provisioned Kubernetes workloads.

  2. Trading a Cloud Identity for Your Own: Workload Attestation on Managed Compute — Netflix TechBlog, September 25 A security-architecture piece on verifying workload identity cryptographically rather than trusting the cloud provider’s identity by default — relevant if zero-trust is on your roadmap.

  3. Taming the ML Firehose: Scaling Feature Consistency — Uber Engineering, September 22 Uber’s approach to keeping ML feature computation consistent between training and serving at massive scale — a recurring architectural headache in any org running production ML.

  4. How Uber Protects Against Retry Storms — Uber Engineering, September 17 A concise resilience-engineering post on the retry-storm failure mode (cascading retries amplifying an outage) and the backoff/circuit-breaking patterns Uber uses to prevent it — core reading for distributed-systems architecture.

  5. Agentic Workflows in Elasticsearch: Pause an AI Agent for Human Approval, Resume 72 Hours Later — Elastic Search Labs, September 21 A concrete pattern for durable, human-in-the-loop agent orchestration — pausing an agent mid-workflow and resuming it days later. Directly applicable if you’re designing approval gates into any agentic system.

  6. Beyond the Model: Engineering AI Infra with Scientific Judgement — Airbnb Tech Blog, September 2026 Airbnb’s “Insight Miner” infrastructure encodes scientific rigor into large-scale, AI-assisted data analysis rather than fully automating it away — a thoughtful take on where human judgment should stay in the loop as AI infra scales.

AI & Engineering Culture

  1. Note on 24th September 2026 — Simon Willison, September 24 A short, provocative note arguing that coding agents are making software engineering harder, not easier, because getting real value out of them demands more discipline and expertise, not less. Good discussion-starter for a team retro on how you’re actually using AI tools.