⚡️ F# Weekly #32, 2026
A new issue of the F# Weekly digest is out. The week's main releases are the FSharp.Data 8.2.0 data-access library and the Mibo 4.0 game framework, whose notably faster PBR pipeline for 3D rendering reaches GPU-bound performance across the raylib and MonoGame backends. The issue also covers fresh updates to FSharp.TypeProviders.Templates 8.11, SQLProvider 1.5.27, SwaggerProvider, linq2db.FSharp, and Polars.FSharp. Notable projects include Plinth (a local-first markdown notebook built with F# and Tauri), Thuja (a minimalist library for terminal UIs), and FsLangMCP (an MCP server that exposes F# compiler services to AI agents). A key theme of the issue is the growth of the F# MCP ecosystem for AI workflows. From Microsoft: NuGet security enhancements, testing-platform improvements, and Agent Framework updates.
♻️ Subscribe for free now!
A new issue of the F# Weekly digest is out. The week's main releases are the FSharp.Data 8.2.0 data-access library and the Mibo 4.0 game framework, whose notably faster PBR pipeline for 3D rendering reaches GPU-bound performance across the raylib and MonoGame backends. The issue also covers fresh updates to FSharp.TypeProviders.Templates 8.11, SQLProvider 1.5.27, SwaggerProvider, linq2db.FSharp, and Polars.FSharp. Notable projects include Plinth (a local-first markdown notebook built with F# and Tauri), Thuja (a minimalist library for terminal UIs), and FsLangMCP (an MCP server that exposes F# compiler services to AI agents). A key theme of the issue is the growth of the F# MCP ecosystem for AI workflows. From Microsoft: NuGet security enhancements, testing-platform improvements, and Agent Framework updates.
♻️ Subscribe for free now!
Sergey Tihon's Blog
F# Weekly #32, 2026 — FSharp.Data 8.2.0, Mibo 4.0, and F# MCP Ecosystem Grows
Welcome to F# Weekly, A roundup of F# content from this past week: Microsoft News Strengthening NuGet Supply Chain Security: Reducing API Key Lifetime Test reporting in Microsoft.Testing.Platform: …
⚡️ Six performance improvements in .NET 11
Steven Giesel breaks down six notable performance improvements in .NET 11, with benchmarks of .NET 10 vs .NET 11 Preview 6 on an Apple M2 Pro. They include: eliminating boxing in generic enum equality checks — up to 86% faster with zero allocations; removing redundant range and null checks in vector operations (~10%); timezone-rule caching that speeds up DateTime.Now by 48%; Guid parsing optimized by 20–25%; faster LINQ Min/Max on small types via vector shuffling (up to 70–75% on bytes and shorts); and overall reductions in generated code size. The key takeaway: the gains come automatically — just upgrade, with no changes to your own code.
♻️ Subscribe for free now!
Steven Giesel breaks down six notable performance improvements in .NET 11, with benchmarks of .NET 10 vs .NET 11 Preview 6 on an Apple M2 Pro. They include: eliminating boxing in generic enum equality checks — up to 86% faster with zero allocations; removing redundant range and null checks in vector operations (~10%); timezone-rule caching that speeds up DateTime.Now by 48%; Guid parsing optimized by 20–25%; faster LINQ Min/Max on small types via vector shuffling (up to 70–75% on bytes and shorts); and overall reductions in generated code size. The key takeaway: the gains come automatically — just upgrade, with no changes to your own code.
♻️ Subscribe for free now!
Steven Giesel
.NET 11 Performance Edition
As always my yearly take on the latest .NET release. In a few months dotnet 11 will be released and I wanted to have a look into some performance metrics.
⚡️ F# Weekly #33, 2026
A new issue of the F# Weekly digest is out. The week's main event is the release of .NET 11 Preview 7 with F# language updates. Also in focus is PaketaBot — a dependency-management bot for Paket projects: written in F#, compiled with Fable, and running directly in GitHub Actions with no hosting required, filling the gap where Dependabot lacks Paket support. Other notable projects include fsdb (a MySQL-compatible database server with wire-protocol support and an FParsec SQL grammar), toro (a deep-learning framework bringing PyTorch semantics to F# via TorchSharp), Farse (JSON parsing on System.Text.Json), and Falco.UnionRoutes (type-safe routing for the Falco web framework). The issue also covers fresh library releases: Ionide.Analyzers 0.16, FSharpLint.Core 0.27.1, and Fable.Logging 1.0.
♻️ Subscribe for free now!
A new issue of the F# Weekly digest is out. The week's main event is the release of .NET 11 Preview 7 with F# language updates. Also in focus is PaketaBot — a dependency-management bot for Paket projects: written in F#, compiled with Fable, and running directly in GitHub Actions with no hosting required, filling the gap where Dependabot lacks Paket support. Other notable projects include fsdb (a MySQL-compatible database server with wire-protocol support and an FParsec SQL grammar), toro (a deep-learning framework bringing PyTorch semantics to F# via TorchSharp), Farse (JSON parsing on System.Text.Json), and Falco.UnionRoutes (type-safe routing for the Falco web framework). The issue also covers fresh library releases: Ionide.Analyzers 0.16, FSharpLint.Core 0.27.1, and Fable.Logging 1.0.
♻️ Subscribe for free now!
Sergey Tihon's Blog
F# Weekly #33, 2026 — .NET 11 Preview 7 Ships with F# Updates and PaketaBot
Welcome to F# Weekly, A roundup of F# content from this past week: News F# in .NET 11 Preview 7 — Release Notes Core Framework v3 4.0.0 [2026 August 14] | xUnit.net Microsoft News .NET 11 Preview 7…
❤1
⚡️ How agent memory works in the Microsoft Agent Framework
Jesse Liberty explains how AI agents in .NET remember what happens to them. Memory is handled by Agent Memory, built on the Neo4j graph database, and it works at three levels: short-term memory holds the current conversation, long-term memory stores accumulated knowledge and the relationships between facts, and "reasoning memory" keeps the agent's past actions and decisions. The author highlights bitemporality in particular: the system records not only when a fact occurred, but also when it recorded that fact. The article also walks through code: connecting to Neo4j, creating a MemoryStore, and the StoreMemory() and RetrieveMemory() methods. This lets an agent keep context across sessions.
♻️ Subscribe for free now!
Jesse Liberty explains how AI agents in .NET remember what happens to them. Memory is handled by Agent Memory, built on the Neo4j graph database, and it works at three levels: short-term memory holds the current conversation, long-term memory stores accumulated knowledge and the relationships between facts, and "reasoning memory" keeps the agent's past actions and decisions. The author highlights bitemporality in particular: the system records not only when a fact occurred, but also when it recorded that fact. The article also walks through code: connecting to Neo4j, creating a MemoryStore, and the StoreMemory() and RetrieveMemory() methods. This lets an agent keep context across sessions.
♻️ Subscribe for free now!
⚡️ An overview of quantum computing programming languages and frameworks in 2026
The Quantum Insider has published an overview of the quantum programming ecosystem in 2026. The author explains how quantum development fundamentally differs from classical: algorithms are probabilistic, measurement collapses the state and prevents free debugging, and today's processors are limited to 50–300 qubits. The overview covers the major frameworks — IBM's Qiskit as the most widely used, Google's Cirq, PennyLane for quantum machine learning, Microsoft's Q# for Azure Quantum, NVIDIA's CUDA-Q, the Amazon Braket SDK, and others. Key trends include the dominance of Python as the common language, senior developers working across several frameworks at once, hardware abstraction layers, and GPU acceleration. The overall takeaway: most real-world applications today are hybrid, combining classical and quantum computing, while the software stack itself is maturing from experiments to practice.
♻️ Subscribe for free now!
The Quantum Insider has published an overview of the quantum programming ecosystem in 2026. The author explains how quantum development fundamentally differs from classical: algorithms are probabilistic, measurement collapses the state and prevents free debugging, and today's processors are limited to 50–300 qubits. The overview covers the major frameworks — IBM's Qiskit as the most widely used, Google's Cirq, PennyLane for quantum machine learning, Microsoft's Q# for Azure Quantum, NVIDIA's CUDA-Q, the Amazon Braket SDK, and others. Key trends include the dominance of Python as the common language, senior developers working across several frameworks at once, hardware abstraction layers, and GPU acceleration. The overall takeaway: most real-world applications today are hybrid, combining classical and quantum computing, while the software stack itself is maturing from experiments to practice.
♻️ Subscribe for free now!
The Quantum Insider
Top Quantum Programming Languages and Frameworks in 2026
A look at the quantum software tools developers use in 2026, from quantum SDKs and compilers to hardware-specific frameworks.
⚡️ Building a distributed job scheduler with Akka.NET
Aaron Stannard of Petabridge describes how to build a distributed scheduler for long-running jobs with Akka.NET — for cases where jobs run from minutes to hours and the cluster must scale without losing work already done. He explains why plain Akka.Cluster.Sharding doesn't fit here: when the cluster resizes, it moves jobs to new nodes and throws away work already completed. Instead, the system is built on several actors, where the main one, JobTracker, keeps the schedule and survives crashes. The key idea is to distribute jobs by nodes' actual load: the scheduler estimates a job's "size" and sends it to the least busy node, avoiding overloading individual machines. This approach wastes no computation and fits well for transcription, transcoding, ETL, and financial reporting.
♻️ Subscribe for free now!
Aaron Stannard of Petabridge describes how to build a distributed scheduler for long-running jobs with Akka.NET — for cases where jobs run from minutes to hours and the cluster must scale without losing work already done. He explains why plain Akka.Cluster.Sharding doesn't fit here: when the cluster resizes, it moves jobs to new nodes and throws away work already completed. Instead, the system is built on several actors, where the main one, JobTracker, keeps the schedule and survives crashes. The key idea is to distribute jobs by nodes' actual load: the scheduler estimates a job's "size" and sends it to the least busy node, avoiding overloading individual machines. This approach wastes no computation and fits well for transcription, transcoding, ETL, and financial reporting.
♻️ Subscribe for free now!
Petabridge
Building a Distributed Job Scheduler with Akka.NET
⚡️ AWS Control Tower vs Azure Landing Zone vs Google Cloud: comparing cloud governance approaches in 2026
Tech Insider compares the three reference architectures for governing multi-account cloud environments in 2026 — AWS Control Tower, Azure Landing Zone built on the Enterprise-Scale accelerator, and Google Cloud's approach based on Organization Policies and Terraform. None of them charges a license fee, yet the real bills differ dramatically. The key takeaway: "free" governance hides the cost of infrastructure — a minimal Azure network hub with Azure Firewall Standard runs about $913 a month before any workloads, while a comparable 10-account setup costs around $150–400 on AWS and $100–300 on Google Cloud. The author also breaks down the guardrail differences: AWS has 750+ prebuilt controls, Azure bets on composable policies via Azure Policy, and Google relies on strict Organization Policies plus detection through Security Command Center. Deployment time, IaC integration, and account structure are compared as well.
♻️ Subscribe for free now!
Tech Insider compares the three reference architectures for governing multi-account cloud environments in 2026 — AWS Control Tower, Azure Landing Zone built on the Enterprise-Scale accelerator, and Google Cloud's approach based on Organization Policies and Terraform. None of them charges a license fee, yet the real bills differ dramatically. The key takeaway: "free" governance hides the cost of infrastructure — a minimal Azure network hub with Azure Firewall Standard runs about $913 a month before any workloads, while a comparable 10-account setup costs around $150–400 on AWS and $100–300 on Google Cloud. The author also breaks down the guardrail differences: AWS has 750+ prebuilt controls, Azure bets on composable policies via Azure Policy, and Google relies on strict Organization Policies plus detection through Security Command Center. Deployment time, IaC integration, and account structure are compared as well.
♻️ Subscribe for free now!
Tech Insider
AWS Control Tower vs Azure Landing Zone vs GCP 2026
AWS Control Tower vs Azure Landing Zone vs Google Cloud landing zones compared: pricing, guardrails, deployment time, and IaC support for 2026.
⚡️ .NET 11 RC1 ships with a go-live license; C# 15 stabilizes union types
.NET 11 Release Candidate 1 is supported in production. C# 15 stabilizes unions and closed hierarchies, System.Text.Json adds union and closed-type polymorphism support, Half uses FP16 hardware instructions. GA lands at .NET Conf, November 10–12.
♻️ Subscribe for free now!
.NET 11 Release Candidate 1 is supported in production. C# 15 stabilizes unions and closed hierarchies, System.Text.Json adds union and closed-type polymorphism support, Half uses FP16 hardware instructions. GA lands at .NET Conf, November 10–12.
♻️ Subscribe for free now!
⚡️ 100 GRPO steps on a free Colab GPU lift a 350M model's structured output score by 7 points
Hugging Face fine-tuned LiquidAI/LFM2.5-350M with GRPO in TRL on ~500 samples and three reward functions: IFStruct rose from 22.6% to 29.7% overall and from 18.0% to 31.9% on JSON, on a 16 GB free-tier Colab or Kaggle GPU with LoRA r=16.
♻️ Subscribe for free now!
Hugging Face fine-tuned LiquidAI/LFM2.5-350M with GRPO in TRL on ~500 samples and three reward functions: IFStruct rose from 22.6% to 29.7% overall and from 18.0% to 31.9% on JSON, on a 16 GB free-tier Colab or Kaggle GPU with LoRA r=16.
♻️ Subscribe for free now!
huggingface.co
Fine-tuning a 350M Model for Better Structured Outputs in 100 GRPO Steps
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
⚡️ Two zones or three? A design framework for zone-resilient Azure workloads
Mark Russinovich and co-authors argue that zone resiliency is a per-component decision, not one number for a whole workload. Three replicas are not three failure domains: a quorum system split across two zones still loses quorum.
♻️ Subscribe for free now!
Mark Russinovich and co-authors argue that zone resiliency is a per-component decision, not one number for a whole workload. Three replicas are not three failure domains: a quorum system split across two zones still loses quorum.
♻️ Subscribe for free now!
Microsoft News
Two zones or three? A design framework for zone-resilient Azure workloads
Learn how to design zone-resilient Azure workloads by choosing the right availability zone strategy for each component, balancing reliability, cost, and operational complexity.
⚡️ Rider's refactoring engine as an agent skill: 157s down to 26s
JetBrains measured a frontier model on fifteen C# refactoring tasks. With the bundled refactoring-code skill, median task time dropped from 157.9s to 26.6s, cost per solved task from USD 0.52 to 0.19, and dotnet build calls from 163 to 3.
♻️ Subscribe for free now!
JetBrains measured a frontier model on fifteen C# refactoring tasks. With the bundled refactoring-code skill, median task time dropped from 157.9s to 26.6s, cost per solved task from USD 0.52 to 0.19, and dotnet build calls from 163 to 3.
♻️ Subscribe for free now!
The JetBrains Blog
Rider Hands AI Agents The Keys To Its Refactoring Engine For Safer, Faster, And Cheaper Results - The JetBrains Blog
We traced a frontier model through fifteen C# refactoring tasks and counted what it reached for. It piped text into interactive commands 468 times. It called git 422 times and sed 392 times. It ran do
⚡️ Rider took 44 s to start because of Defender — Microsoft fixed it in Defender itself
JetBrains traced slow Rider and ReSharper startup to Microsoft Defender scanning processes installed in user-writable folders: up to 44 seconds versus under one second for Visual Studio. Defender 1.449.454.0 brings it under 5 seconds.
♻️ Subscribe for free now!
JetBrains traced slow Rider and ReSharper startup to Microsoft Defender scanning processes installed in user-writable folders: up to 44 seconds versus under one second for Visual Studio. Defender 1.449.454.0 brings it under 5 seconds.
♻️ Subscribe for free now!
The JetBrains Blog
Why Rider and ReSharper Were Slow to Start, and How Microsoft Helped Fix the Problem - The JetBrains Blog
Find out how we teamed up with Microsoft to optimize startup times of ReSharper and JetBrains Rider with Microsoft Defender for Windows Developers.
⚡️ Azure and AWS launch Multicloud Interconnect: private connectivity up to 100 Gbps
Microsoft and AWS jointly announced Azure Multicloud Interconnect for AWS — standardized private connectivity between the two clouds, up to 100 Gbps at GA, integrated with Azure Private Link. Listed as Public Preview in Azure Updates.
♻️ Subscribe for free now!
Microsoft and AWS jointly announced Azure Multicloud Interconnect for AWS — standardized private connectivity between the two clouds, up to 100 Gbps at GA, integrated with Azure Private Link. Listed as Public Preview in Azure Updates.
♻️ Subscribe for free now!
Microsoft News
Introducing Azure Multicloud Interconnect for AWS
Learn how Azure Multicloud Interconnect helps organizations simplify Azure and AWS connectivity, support AI workloads, and build scalable multicloud architectures through an open, cloud-native networking experience.
⚡️ PyTorch 2.14: NVGEMM backend, nccl2 and fault tolerance as a first-class c10d concept
PyTorch 2.14 lands with 2,995 commits from 487 contributors: the CuTeDSL-based NVGEMM backend, an in-tree nccl2 distributed backend with one-sided RMA windows, in-place process-group reconfiguration, @dynamic_spec and torch.switch in torch.compile, and a rewritten MPS linear algebra stack. Python 3.15 is eager-only.
♻️ Subscribe for free now!
PyTorch 2.14 lands with 2,995 commits from 487 contributors: the CuTeDSL-based NVGEMM backend, an in-tree nccl2 distributed backend with one-sided RMA windows, in-place process-group reconfiguration, @dynamic_spec and torch.switch in torch.compile, and a rewritten MPS linear algebra stack. Python 3.15 is eager-only.
♻️ Subscribe for free now!
⚡️ .NET September 2026 servicing closes 8 CVEs across 10.0, 9.0 and 8.0
The September 8 servicing release ships .NET 10.0.12, 9.0.20 and 8.0.31 and fixes eight CVEs. One of them, CVE-2026-69522, also affects .NET Framework 4.6.2 through 4.8.1.
♻️ Subscribe for free now!
The September 8 servicing release ships .NET 10.0.12, 9.0.20 and 8.0.31 and fixes eight CVEs. One of them, CVE-2026-69522, also affects .NET Framework 4.6.2 through 4.8.1.
♻️ Subscribe for free now!
⚡️ Rider 2026.3 EAP 1: rainbow brackets, data breakpoints from the editor, completion filters
The first Rider 2026.3 EAP build adds rainbow brackets (off by default), data breakpoints set by right-clicking a variable, filters in the completion popup, a Game Development plugin category, and dotCover support for TUnit.
♻️ Subscribe for free now!
The first Rider 2026.3 EAP build adds rainbow brackets (off by default), data breakpoints set by right-clicking a variable, filters in the completion popup, a Game Development plugin category, and dotCover support for TUnit.
♻️ Subscribe for free now!
The JetBrains Blog
The Rider 2026.3 Early Access Program Is Open - The JetBrains Blog
The first Early Access build for Rider 2026.3 is now available! It includes rainbow brackets, an easier way to set data breakpoints, a new Game Development plugin category, and filters in code complet
⚡️ Azure Connector Namespace hosts MCP servers for Azure SRE Agent, no infrastructure to run
Azure Connector Namespace is a managed service for hosting connectors and MCP servers — deployment, HTTPS, auth, secrets and scaling handled by the platform. The Azure SDK team shows how to deploy an Azure SQL MCP server with azd up and plug it into Azure SRE Agent via managed identity.
♻️ Subscribe for free now!
Azure Connector Namespace is a managed service for hosting connectors and MCP servers — deployment, HTTPS, auth, secrets and scaling handled by the platform. The Azure SDK team shows how to deploy an Azure SQL MCP server with azd up and plug it into Azure SRE Agent via managed identity.
♻️ Subscribe for free now!
⚡️ funes gives coding agents a memory built from their own session traces
Hugging Face released funes, a local memory layer for Claude Code, Codex, pi and Hermes. It indexes session traces into a Lance dataset; recall returns original passages with provenance and can sync to a private Hub dataset.
♻️ Subscribe for free now!
Hugging Face released funes, a local memory layer for Claude Code, Codex, pi and Hermes. It indexes session traces into a Lance dataset; recall returns original passages with provenance and can sync to a private Hub dataset.
♻️ Subscribe for free now!
huggingface.co
Give Your Coding Agents a Memory You Own
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
⚡️ Uno Platform split their MCP server in two by data lifetime, not by feature
A stateless docs server and a stateful local app server cost the agent 6.4k and 1.5k context tokens respectively, versus 5.2k for GitHub's single built-in MCP server.
♻️ Subscribe for free now!
A stateless docs server and a stateful local app server cost the agent 6.4k and 1.5k context tokens respectively, versus 5.2k for GitHub's single built-in MCP server.
♻️ Subscribe for free now!
⚡️ Azure per-disk resiliency preview keeps a VM running when one data disk drops
With per-disk resiliency, Azure takes only the affected data disk offline and reattaches it once connectivity returns, instead of recovering the whole VM. Announced alongside Azure Infrastructure Resiliency Manager.
♻️ Subscribe for free now!
With per-disk resiliency, Azure takes only the affected data disk offline and reattaches it once connectivity returns, instead of recovering the whole VM. Announced alongside Azure Infrastructure Resiliency Manager.
♻️ Subscribe for free now!
Microsoft News
The future of infrastructure resiliency starts with modernization
Learn how Azure can help assess resiliency posture, identify gaps, and increase uptime with recommendations, operational guidance, and more.
⚡️ IBM's Granite PatchTST-FM-r2: SOTA zero-shot time series forecasting under Apache 2.0
The ~385M-parameter model ranks second for CRPS and MASE among replicable zero-shot models on GIFT-Eval and first among permissively licensed ones. Conformer blocks, 8,192-step context, 99 quantiles, Apache-2.0 and OpenMDW-1.0.
♻️ Subscribe for free now!
The ~385M-parameter model ranks second for CRPS and MASE among replicable zero-shot models on GIFT-Eval and first among permissively licensed ones. Conformer blocks, 8,192-step context, 99 quantiles, Apache-2.0 and OpenMDW-1.0.
♻️ Subscribe for free now!
huggingface.co
IBM releases SOTA Granite Time Series PatchTST-FM-r2 model with commercial-friendly license
A Blog post by IBM Research on Hugging Face