
Programming Notes November 12, 2026
- Programming
- Cloud
- AI Engineer
As usual, I’ll introduce a carefully curated selection of tech news and articles I found online today that personally caught my attention. I’d be happy to hear your opinions and any additional information in the comments.
Programming
Go
The Microsoft builds of the Go language, “v1.27.4-1” and “v1.26.11-1,” which include security fixes, have been released and are now available for download.
The “agent harness” (agent/harness package) added to the Microsoft Agent Framework for Go is a preconfigured mechanism that coordinates tool execution and provides context between model calls.
By using the harness, task decomposition prompts, a ToDo management tool, planning and execution modes, FileMemory for retaining memories on disk, conversation history management, and other components are configured automatically. Furthermore, the initial settings for various features—such as file-based skill loading, tool execution approval rules, hosted web search, message insertion during execution, and tracing with OpenTelemetry—can all be arranged at once.
Information about ToDo items and working folders within a session is held in the process’s memory, while data such as plans is saved to a designated directory under the working directory. By serializing, saving, and restoring the session state, work can be resumed even after a restart.
A 2026 survey found that many Go developers are experienced engineers and that full-stack development combining Go with TypeScript, in addition to backend development, is rapidly increasing. The use of the standard library is expanding and has reached a level nearly on par with Gin.
Also, the use of AI agents centered on CLIs is extremely active, but final code reviews tend to remain the responsibility of humans. In terms of the language itself, generic methods, a new GC, and other features were introduced in Go 1.26 and 1.27, bringing about major advances.
PowerShell
The general availability (GA) of PowerShell 7.7 has been announced. This version is based on .NET 11 and is an STS (Standard Term Support) release focused on strengthening security, adding new features, and improving reliability.
New features such as commands for creating temporary directories have been added, while some breaking changes, such as changes to the default GUID generation behavior, are also included. In addition, with the deprecation of MSI packages, migration to the MSIX format is underway.
JetBrains Rider
Rider 2026.3 added official plugins for Lua and Verse.
Integration with Godot has also been enhanced, C++ debugging has been accelerated, and CoreCLR support for Unity has been added, improving the game development environment.
Visual Studio Code
Visual Studio Code 1.146 introduced numerous feature additions and improvements centered around agent functionality to enhance work efficiency.
Notable new features include “Inbox mode,” which enables centralized management of session status across all projects, and “Chat windows,” which allow chats to be detached into separate windows for freely adjustable layouts. It is now also possible to search for and install extensions directly from within the Agents window.
The model comparison feature supports Copilot Fusion and can perform automated evaluations by agents.
Expandability and security have also been enhanced through features such as providing skills via MCP, supporting virtual repositories, and masking environment variables to protect sensitive information.
The legacy Local harness has been deprecated and is scheduled for removal in February 2027, so users are being encouraged to migrate to the Copilot harness, which is now the default.
Visual Studio
Visual Studio 18.12 has made “Bring Your Own Model (BYOM)” generally available, enabling flexible integration and use of self-hosted or local AI models without signing in to GitHub.
In ReSharper 2026.3, Claude Agent and Codex are now available alongside the existing Junie in the AI Assistant window. These agents can perform ReSharper’s refactoring operations, and a free tier of Junie Lite is also available.
It also supports new syntax from day one, including union types in .NET 11 and C# 15. In addition, development support features have been extensively updated, including enhanced code analysis, improved decompilation, and test runner support for xUnit v3 4.0 and NUnit 5.0.
SQL Server
In 2026, Microsoft significantly expanded AI integration and cloud capabilities across its SQL-related services.
Azure SQL Database Hyperscale introduced large-scale compute provisioning and enhanced vector search capabilities, advancing the integration of operational data with AI.
A wide range of updates has also been released to improve development and operational efficiency, including integration of agent capabilities into SSMS and Visual Studio Code, Microsoft Fabric integration, and enhanced migration support tools.
This article introduces the overview and learning resources for a new certification, “DP-800 (SQL AI Developer Associate),” which is designed for building and optimizing AI-enabled solutions by leveraging SQL database development skills.
Kiro
The Kiro AI development environment has introduced a new “Approve for me” feature. An LLM reviews operations not defined by permission rules, automatically approves safe routine operations, and blocks dangerous operations. Users are asked to confirm only when a difficult judgment is required.
This mechanism significantly reduces the effort required for approval while maintaining security.
GitHub / GitHub Copilot
Google’s latest model, “Gemini 4 Carbon,” has reached general availability (GA) in GitHub Copilot, allowing users on eligible paid plans to use it for security-focused coding and reasoning tasks from various development environments.
GitHub’s dashboard has been updated with features such as pinning the default display tab, reducing whitespace, and increasing the default number of displayed items, making it easier to view more information at a glance.
By using the fine-grained permissions added for enterprises, Enterprise GitHub Apps can call the enterprise-level Copilot Sheets management API and control the assignment of Copilot Sheets to users and enterprise teams.
It is now also possible to configure issue types and custom fields, as well as directly create labels and milestones from the issue screen, in repositories belonging to personal GitHub accounts.
Agentic Coding and Specification-Driven Development
This is a case study of an AI-driven development (AI-DLC) workshop conducted by MBK Wellness and AWS. A total of 22 participants, including executives, experienced requirements definition and mockup creation for a real-world project while practicing a development process in which “AI takes the lead and people make the decisions.”
As a result, it became clear that the true bottleneck to adopting AI was not technology but interdepartmental coordination and the decision-making process. The company is now working to delegate authority and flatten its organization to maximize development speed.
AI
Microsoft
The GPT-5.6, GPT-6, and GPT-6.1 families have been added to Azure Content Understanding, along with guidelines for selecting the optimal model based on workload quality and token cost requirements, as well as evaluation data.
Microsoft Foundry has added Anthropic’s latest models, “Claude Fable 5.2” and “Claude Mythos 5.2,” enabling advanced professional work in areas such as software engineering, life sciences, and security.
Gemini Intelligence is not a single feature name, but rather an umbrella term for a suite of capabilities tailored to seven specific models that meet high system standards. It is distinct from automated features delivered through apps, and its availability is also expanding gradually in Japan.
Google announced a new endpoint, “Gemini Decisions API,” for fast multimodal decision-making, powered by the new “Gemini Decision 1” model. It can simultaneously evaluate up to 1,000 independent questions in a single operation, with inputs of up to 256K tokens in formats including text, images, audio, video, and PDFs.
Unlike conventional methods that generate text one token at a time, it scores options in a single forward pass, delivering end-to-end latency as low as 150–250 milliseconds. It supports three evaluation formats: “yes/no (noul),” “multiple choice (choice),” and “rubric evaluation (score),” with every response assigned a confidence score from 0 to 1.
Pricing is $0.10 per million input tokens, with no charge for output tokens. Google AI Studio offers a free tier for prototyping, and the API is already being used within Google for capabilities such as model routing and safety measures.
Google DeepMind announced its latest vision-language-action (VLA) model, “Gemini Robotics 2.5.” The model significantly improves dexterity, reliability, and safety, accelerating the learning and execution of diverse real-world tasks on factory production lines and in everyday environments.
Google DeepMind announced “AlphaGenome 2,” its next-generation DNA sequence model. It accurately predicts, at the single-cell level, the effects that specific genetic mutations have on particular cell types or disease states, helping advance the understanding of disease mechanisms and drug discovery.
The Gemini app is now available in Apple CarPlay, enabling natural, hands-free conversations while driving to check schedules, summarize emails, brainstorm ideas, and more.
JetBrains
JetBrains IDE’s 2026.3 release introduced “Air,” a workspace for centrally managing AI coding agents.
It supports running multiple agents, such as Junie and Claude, in parallel, separating work across Git worktrees, and reviewing changed sections. The free Junie Lite is also available, allowing developers to use AI efficiently while remaining in control.
Papers and Other
Google released “A2UI 1.0,” an open protocol that enables AI agents to generate interactive screens, not just text.Unlike conventional text-centric responses, it can present interactive interfaces such as forms and graphs, enabling intuitive handling of complex schedule coordination and information comparison.
In A2UI, agents describe the UI composition, and each application, such as Web, iOS, and Android, renders it safely and quickly using its own native components. Renderers such as React and Flutter, as well as developer SDKs for languages such as Python and TypeScript, are also provided, enabling a consistent interaction experience even within AI assistants such as Gemini and Claude.
Companies such as Atlassian, ByteDance, and Alation have already begun adopting it, and it is also being used within Google, including in Gemini Notebook. It is expected to become a new foundation for easily building flexible interactive interfaces tailored to users’ situations across platforms.
This article addresses the problem of coding agents reimplementing the same functionality without using existing functions. Rather than providing the full source code, you record a concise function list in an instruction file and pass it along, summarizing each function’s “location, name, arguments and return value, and behavior,” which greatly improves code reuse. It also states that checking for duplication during reviews is important.
The article explains, based on interviews with the paper’s authors and independent replication results, that when major LLMs are asked cultural questions without specifying a location, a bias toward citing Japan as an example persists even in the latest models, and that preventing this bias requires the prompt side to explicitly assign countries programmatically.
Cloud
Azure
Azure Updates (2026-11-12) | Buchizakki
A summary of the latest Microsoft Azure and .NET-related updates as of November 12, 2026.
The biggest topic is the official release of .NET 11. As a result, general availability (GA) of .NET 11 has also begun for Azure Functions. Meanwhile, support for .NET 8 and .NET 9 has ended for Azure Static Web Apps, so early migration is recommended.
For AI and Microsoft Foundry, support has been added for the latest models, including Anthropic’s “Claude Fable 5.2 / Mythos 5.2” and Moonshot AI’s “Kimi K3.1 Pro,” as well as for media-generation models provided by fal. Real-time multilingual transcription and multichannel speech recognition in Speech API have also become generally available (GA).
Many features have also been added to other major services. Work is underway to support 64-bit on the Windows Free plan in App Service and prepare TLS for post-quantum cryptography (PQC), while edge actions and post-quantum key exchange policies have reached GA in Azure Front Door. Distributed Accelerators for Cloud Storage (DACS) for AI workloads on AKS have also been released in preview, and a wide range of updates have been implemented across infrastructure and developer tools in general.
Support for .NET 11 in the isolated worker model is now generally available (GA) in Azure Functions. It currently supports C#.
This release introduces a new project SDK and simplifies configuration. When migrating or updating existing applications that use the in-process model or the Linux Consumption plan, supported environments and compatibility should be checked in advance.
Cloudflare
Cloudflare has adopted the new consensus algorithm “QuePaxa” in its distributed service “Meerkat,” eliminating the need for leader dependency or failure detection so that systems can continue operating without interruption even in unstable global network environments.
OS
Windows
Microsoft released its monthly security updates on November 10, 2026 (U.S. time). Critical vulnerabilities, including remote code execution vulnerabilities, have been fixed in products such as Windows and Office.
In particular, the Microsoft Outlook spoofing vulnerability was confirmed to have been exploited before disclosure, prompting calls for prompt application of the fix. Automatic updates are enabled in many environments, but users of the affected products are advised to apply the updates early or review the details.
macOS
Canadian company Tunabelly Software released “TG Pro v2.105,” a system monitoring app that supports fan control and temperature sensors for MacBook Pro and iMac models equipped with the Apple M6 chip, as well as MacBook Pro models equipped with a PROLED XDR display.
Microsoft released the “Microsoft 365/Office for Mac v16.115” security update on November 10, 2026.
A total of 140 vulnerabilities, the highest number ever recorded for a single month, were fixed in products such as Excel, Word, and PowerPoint. In particular, the fixes for products such as Excel and Word include multiple remote code execution (RCE) vulnerabilities, so their prompt application is recommended.
Microsoft used AI (Vibe Coding) to port and release the popular Windows process management tool “Sysinternals Process Explorer” for Mac.
It supports macOS 14 and later and includes core features such as a real-time process tree display and system information graphs.
Industry Trends and Current Affairs
The internet continues to serve as humanity’s greatest archive, preserving traces of the dead, but its data are managed according to corporate profitability and are therefore extremely vulnerable.
In recent years, the proliferation of low-quality AI-generated content and the increasing closedness of platforms have made the cultural decline and demise of the web increasingly realistic topics of discussion.
However, even such corruption and decline may become “compost” for creating a better system to come.Go to list of users who liked
Go to list of comments
Register as a new user and use Qiita more conveniently
- You get articles that match your needs
- You can efficiently read back useful information
- You can use dark theme



