* * 2025 Annual Summary for. NET Developers **

I believe that this year you have seen many articles like "Sorry, C#has fallen out of the first echelon". What is the. NET ecosystem? This article will systematically sort out the technology trends and important events that. NET developers should pay attention to in 2025, covering the latest developments and trends in AI development,. NET evolution and the integration of the two, in order to help everyone find the right position and meet future challenges and opportunities.

---

This article was originally written by me (Shengjie) and completed with AI assistance.

This article only states facts and expresses personal opinions, and does not sell anxiety. Please read it with confidence. The article is long, so it is recommended to like and collect it first.

Part of the content of this article refers to public information such as Microsoft's official blog,. NET blog, NuGet statistics, and TIOBE Programming Language Index. There may be omissions. Please correct them.

* * Me in 2025? *🎯

Before you start, take a few seconds to review your 2025:

• []** Keep up with the trend and apply what you have learned **-I learned a lot of new technologies and used them at work📚
• []** Embracing AI, doubles efficiency **-Vibe Coding is so fragrant, it can no longer be separated from AI assistants🚀
• [] Deiled ** Full harvest, higher level **-Promotion and salary increase/Successful job-hopping/Project sold well🎉
• []** AI is anxious, where should I go? *-Worried about being replaced by AI, I don't know what to learn😰
• []** Tired on the run, standing still **-Busy with business and no time to learn new technologies😩
• []** Transformation exploration, ready to go **-Transforming or exploring new directions🌱
• []** The Buddhist system lies flat, the years are quiet **-It's good to be stable, I don't want to curl up anymore😎

No matter which type you are, this article hopes to bring you some inspiration and direction.

• • If possible, please share your thoughts on 2025 and expectations for 2026 in the comment area **

---

* * Preface **

2025 is destined to be a colorful year in the history of technological development. Looking back on this year, AI Agents swept the world,. NET and AI were deeply integrated, and developer tools blossomed in bloom. As a. NET developer, are you ready for this change?

If 2023 is the first year of large models and 2024 is the first year of the implementation of large models, then 2025 is undoubtedly the first year of AI Agent. From simple code completion to intelligent agents that can independently plan, reason, and invoke tools, AI's capabilities have made a qualitative leap.

At the same time, the. NET ecosystem has also ushered in an important milestone-. NET 10 was officially released, the C#language continues to climb on the TIOBE list, NuGet weekly downloads exceeded 5.5 billion, and Visual Studio 2026 brings a new experience. What is even more exciting is that Microsoft continues to make efforts in the field of AI. Frameworks such as Microsoft. Extensions. AI, Semantic Kernel, and Microsoft Agent Framework are becoming increasingly mature.. NET developers have a complete tool chain for building AI applications.

This article will sort out the technical trends and important events that. NET developers should pay most attention to in 2025 from three dimensions: AI development,. NET evolution, and. NET + AI integration.

---

* * 1. Development of AI **

* * 1.1 The strongest AI model in history is constantly refreshed **

In 2025, AI model capabilities will advance by leaps and bounds, with major manufacturers chasing each other and constantly setting new performance records.

* * Foreign models **

manufacturers	model	Highlights/Features
OpenAI	GPT-5.2	Flagship universal model, reasoning and tool invocation have been significantly improved
Anthropic	Claude Opus 4.5	Flagship, best performance in programming and agent, high efficiency in Token
Google	Gemini 3 Pro	Flagship model to enhance the Deep Research and Gemini App experience
XAI	Grok 4.1	Superior emotional intelligence

* * Domestic model **

manufacturers	model	Highlights/Features
* * Alibaba Cloud **	Qwen3	The new flagship supports 119 languages, greatly improves reasoning and Agent capabilities, and the open source version leads the domestic model ecosystem.
* * Intelligent AI **	GLM-4.7	New generation flagship, fully upgraded programming/Agent/reasoning/dialogue
DeepSeek	DeepSeek-V3.2	Flagship model, enhanced agent capabilities, leading cost performance
* * Xiaomi **	Mimo	New flagship model with code capabilities that exceed all open source models

* * Impact on developers **

The improvement of model capabilities directly benefits developers:

• • • Stronger code generation **: More reliable complex refactoring, architectural design, and bug fixes
• • • Longer context **: Dealing with large codebases and long documents is no longer a problem
• • • Better tool invocation **: Agent capabilities have been greatly enhanced and the degree of automation is higher
• • • Lower costs **: With the same capabilities, API call costs continue to decline

* * 1.2 The agent era is coming **

In 2025, AI will evolve from an "assistant" to an "agent."

In the past, AI assistants (such as the early Copilot) mainly played the role of "smart completion"-you write one line of code, and it helps you complete the next line. Today's AI Agents are completely different. They have four core capabilities:

• • • Planning **: Ability to break down complex tasks into multiple sub-tasks and formulate execution plans
• • • Reasoning **: Ability to conduct logical reasoning based on context, make judgments and decisions
• • • Tool Use **: Ability to call external APIs, execute code, and operate file systems
• • • Memory **: Ability to remember historical conversations and operations, maintaining context during long-term tasks

This means that AI is no longer just "answering questions" but can "complete tasks." You can tell AI to "help me refactor the code of this module." It will automatically analyze the code structure, identify problems, formulate a refactoring plan, perform modifications, run tests, and finally deliver a complete result.

GitHub Copilot's Agent model is a typical example of this trend. When you use Agent mode in VS Code, Copilot no longer just completes code, but can understand your intentions, proactively search the code base, modify multiple files, and run commands until the task is completed.

* * 1.3 Development of the protocol: Building AI interoperability standards **

The explosion of Agent capabilities is inseparable from the promotion of standardized protocols. In 2025, the four major agreements jointly built the infrastructure for AI interoperability.

1. MCP Protocol (Model Context Protocol)

MCP was initiated by Anthropic and has become the de facto standard for AI tool invocation.

Before the advent of MCP, every AI application required dedicated integration code to be written for different tools. MCP unifies the way AI models interact with external tools, defining:

• • • Resources **: Resources that AI can read (files, databases, API responses, etc.)
• • • Tools **: Functions that AI can call (search, calculation, operation, etc.)
• • • Tips **: Pre-defined prompt template

With MCP, developers only need to write an MCP Server once to allow all AI clients that support MCP to use this tool. At present, mainstream AI tools such as Claude, VS Code Copy, and Cursor already support the MCP protocol.

2. A2A (Agent-to-Agent)

A2A was initiated by Google and defines communication standards between agents.

In complex business scenarios, a single agent often cannot complete all tasks. For example, a "travel planning agent" may need to call multiple professional agents such as "flight booking agent","hotel booking agent", and "itinerary planning agent" to work together.

The A2A agreement solves this problem. It defines:

• • • Agent Card **: describes the agent's capabilities and interfaces
• • • Task **: Task requests and responses passed between Agents
• • • Message **: Communication message format between Agents

Through A2A, Agents developed by different teams can collaborate seamlessly to build a more powerful multi-agent system.

3. AG-UI (Agent User Interaction Protocol)

AG-UI defines the interaction protocol between the agent and the user interface.

Traditional AI applications usually have a "question-and-answer" model, but when Agents perform complex tasks, users need to understand progress in real time, view intermediate results, and provide feedback. AG-UI protocol support:

• • • Streaming UI update **: Display the agent's thinking process and execution status in real time
• • • Progress feedback **: Display the task completion percentage and estimated remaining time
• • • Interactive confirmation **: Continue after critical nodes request user confirmation

The Microsoft Agent Framework already has built-in AG-UI support, allowing developers to easily build user-friendly Agent applications.

4. Agent Skills

Agent Skills is an open standard format initiated by Anthropic to give AI Agents new capabilities and expertise.

Unlike other protocols, Agent Skills is not a communication protocol, but a knowledge packaging format. It allows Agents to load procedural knowledge and specific contexts (company level, team level, user level) on-demand to complete tasks more accurately and efficiently.

Agent Skills can achieve:

• • • Domain expertise **: Package expertise into reusable instructions (e.g., legal review processes, data analysis pipelines)
• • • New capabilities **: Give agents new capabilities (such as creating PPT, building MCP Server, analyzing datasets)
• • • Repeatable workflows **: Convert multi-step tasks into consistent and auditable workflows
• • • Interoperability **: The same skill can be reused in different Agent products

Currently, Agent Skills has been adopted by many mainstream AI tools, including GitHub Copilot, VS Code, Claude Code, Cursor, OpenAI Codex, Amp, Goose, etc.

It is worth noting that. NET 10 's new ability to directly run. cs files makes writing scripts in Agent Skills easier-you can write tool scripts directly in C#without having to create a complete project.

* * 1.4 AI IDE rolls to smoke **

In 2025, the competition for AI programming tools will enter a heated stage. ** Have you turned on Vibe Coding mode? **

* * Foreign products **

products	manufacturers	Highlights/Features
GitHub Copilot	GitHub	Market leader, Agent model supports autonomous tasks, deeply integrates VS Code/VS
Claude Code	Anthropic	Terminal assistant, good at handling large code bases, based on the Claude model
Codex CLI	OpenAI	Command-line tool, lightweight and fast, suitable for terminal workflows
Cursor	Cursor	AI-first IDE representative, smooth experience, support multiple model switching
Windsurf	Codeium	AI IDE focuses on the concept of "Flow" and emphasizes human-computer collaboration
Kiro	AWS	Deep integration with AWS, suitable for cloud-native development

* * Domestic products **

Domestic manufacturers are also unwilling to lag behind and have launched their own AI programming tools:

products	manufacturers	Highlights/Features
CodeBuddy	腾讯云	Support for multiple languages and frameworks
TRAE	字节跳动	Based on the big bean bag model
Qoder	Alibaba	Deep integration with Alibaba Cloud

For. NET developers, there are many options, but among many AI programming tools, GitHub Copilot ** is still the first choice-it has the most mature integration with Visual Studio and VS Code, and has the best support for C#.

---

* * 2. Development of. NET **

* * 2.1. NET 10 released **

In November 2025,. NET 10 will be officially released, which is another important milestone for the. NET platform.

* * C#language status improved **

According to the TIOBE Programming Language Index, C#will continue to climb in 2025 and is expected to become the language of the year. This is due to:

• Continued enhancement of. NET cross-platform capabilities
• Unity game development continues to be popular
• Stable needs for enterprise-level application development
• Improvement of the. NET ecosystem in AI/ML field

* *. NET 10 Core Highlights **

NET 10 continues the tradition of performance first and brings several important improvements:

• • • Continuous performance optimization **: Improved JIT compiler, lower memory footprint
• • • Enhanced container support **: Smaller image size, faster boot speed
• • • Native AOT improvements **: Support more scenarios and improve compilation speed
• • • Cloud-native enhancements **: Better integration with Kubernetes, container-based deployments

* * NuGet Ecological Prosperity **

NuGet's weekly downloads exceeded 5.5 billion, a figure that fully demonstrates the activity of the. NET ecosystem. From web development to machine learning, from game development to the Internet of Things, you can find almost all the libraries you need on NuGet.

* *. NET 10 supports running. cs files directly **

This is one of the most exciting features of. NET 10. Now you can run a single. cs file directly:

dotnet run app.cs

不需要创建项目文件，不需要Program.cs的样板代码，就像运行Python脚本一样简单。这极大降低了.NET的入门门槛，也让编写快速原型和工具脚本变得更加方便。

* * DNx debuts: The "npx/uvx" era of. NET **

dnx（.NET eXperience）是.NET 10 SDK引入的全新工具执行脚本，本质上是dotnet tool exec命令的精简、用户友好的包装器。它标志着.NET正式进入"一次性执行"（one-shot）时代，与Python的uvx和Node.js的npx完全对标。

* * Core Features **

• • • Run without installation **: Run the. NET tools directly from the NuGet package without permanent local or global installation
• • • Isolated execution environment **: The toolkit is temporarily downloaded into NuGet cache for execution without modifying the system PATH environment variable to ensure clean isolation
• 智能版本管理：默认使用指定工具包的最新版本，支持@版本号指定版本；优先使用本地.config/dotnet-tools.json中的配置
• • • Smooth development experience **: Streamline workflow and reduce the threshold for trying new tools

* * Usage example **

# 执行C#代码片段
dnx dotnet-execute 'WriteLine("Hello dnx!!!");' --using "static System.Console"

# 快速生成GUID
dnx dotnet-execute "Guid.NewGuid()" 

# 性能压测
dnx LoadTestToolbox hammer --url https://www.example.com --min 1 --max 100

# JSON转YAML
dnx json2yaml -i:input.json -c

# 解码JWT Token
dnx dotnet-decode-jwt <token>

* * Comparison with other platforms **

characteristics	dnx	uvx	npx
Execution on demand	✓	✓	✓
version control	@Version number	@Version number	@Version number
isolated execution	Independent NuGet Cache	temporary virtual environment	temporarily download
ecological maturity	developing	Unified uv tool chain	Deep maturity

This innovation takes the NuGet ecosystem to new heights and opens up new possibilities for developer tool distribution.

* * NuGet package supports distribution of MCP Server **

This is an important step in the integration of. NET and AI ecosystems. Now you can package MCP Server as a NuGet package for distribution. Users only need to install the NuGet package to gain AI tool capabilities. This makes it easier for the. NET community to share AI tools.

* * 2.2 C#14 New Features **

C#14, released with. NET 10, brings a number of practical language improvements.

* * field keyword **

Backup fields can now be accessed directly in automatic properties:

public class Person
{
    public string Name
    {
        get => field;
        set => field = value?.Trim() ?? throw new ArgumentNullException();
    }
}

There is no longer the need to explicitly declare private fields, making the code simpler.

* * Extension Members **

C#14 has greatly enhanced the ability to extend methods, and now you can define extended attributes, extend static members, etc.:

public extension IntExtensions for int
{
    public bool IsEven => this % 2 == 0;
    public static int Zero => 0;
}

// 使用
int x = 10;
bool even = x.IsEven;  // true
int zero = int.Zero;   // 0

* * Params collection enhancement **

params关键字现在支持更多集合类型，不仅限于数组：

void PrintAll(params IEnumerable<string> items)
{
    foreach (var item in items)
        Console.WriteLine(item);
}

// 可以传入List、Array等任何IEnumerable<string>
PrintAll(["a", "b", "c"]);
PrintAll(new List<string> { "x", "y" });

* * 2.3 Framework Ecological Update **

* * Aspire 13 released **

NET Aspire is a cloud-native application development framework launched by Microsoft. In 2025, it will usher in a major upgrade-Aspire 13. This is the largest release to date:

• aspire do 命令：全新的构建、发布、部署流水线体验，支持自定义流水线步骤
• • • Aspire MCP Server **: Dashboard has built-in MCP server, and AI assistant can directly query running application logs and tracking data
• • • Multi-language connection string **: Database resources are automatically exposed in multiple formats (. NET format, Python URI format, Java JDBC format)
• JavaScript/Python一等公民支持：AddJavaScriptApp、AddPythonApp等全新API
• • • New official website www.example.com **: Comprehensive migration of documents and resources

ASP.NET Core 10

• Blazor增强：[PersistentState]声明式状态持久化、电路状态保持、嵌套表单验证、Passkey无密码认证、NotFound页面参数、JS互操作增强
• Minimal API改进：内置AddValidation()验证支持、TypedResults.ServerSentEvents原生SSE、Record类型验证
• • • OpenAPI 3.1 native support **: default generation of 3.1 specification, YAML format output, automatic integration of XML document comments
• • • Certification and authorization indicators **: New OpenTelemetry certification/authorization indicators, and API endpoints intelligently return 401/403

MAUI 10

• • *. NET Aspire integration **: Added project templates to support OpenTelemetry telemetry and service discovery
• • • XAML Source Generator **: Generates strongly typed code at compile time, and simplifies declarations in global XML namespaces
• • • Control enhancements **: HybridWebView supports request interception, CollectionView performance optimization, and MediaPicker multiple selection
• • • Platform improvements **: Android CoreCLR experimental support, iOS binding projects can be built on Windows

EF Core 10

EF Core 10 was released as an LTS version along with. NET 10, bringing a number of important updates:

• 向量搜索支持：完整支持SQL Server 2025和Azure SQL的vector数据类型，通过SqlVector<float>和VectorDistance()函数实现AI语义搜索和RAG场景
• JSON数据类型：原生支持SQL Server 2025的json列类型，查询效率大幅提升
• • • LeftJoin/RightJoin operator **: Supports the new LINQ left and right connection method added to. NET 10, simplifying complex queries
• • • Named query filters **: Support the configuration of multiple named filters for entities, which can be selectively disabled during queries
• • • Complex type enhancement **: Support optional complex types, JSON mapping, and structure mapping
• • • ExecuteUpdate supports JSON columns **: It can efficiently update JSON document attributes in batches
• • • Improved parametric sets **: New default translation mode, where each set value is converted into a separate parameter to optimize query plan caching
• • • Cosmos DB full-text search **: Supports full-text search and hybrid search (RRF function)

* * ABP 10 released **

As.NET's excellent enterprise-level application framework, ABP released version 10.0 in 2025, which is a major update:

• • • Upgrade to. NET 10 **: Full support for the latest. NET 10 LTS versions
• • • AI integration (Volo. Abp. AI)**: Provides unified integration of AI capabilities, supports Microsoft.Extensions.AI, Microsoft Agent Framework and Semantic Kernel, and introduces the AI Workspace concept to achieve isolated configuration
• • • New Workflow module **: Integrate with Elsa Workflows to support building visual, long-running event-driven workflows
• • • Mapply replaces AutoMapper **: Uses compile-time source generator for better performance and requires no run-time reflection

* * 2.4 Visual Studio 2026 released **

One of the most exciting news for. NET developers in 2025 is the release of Visual Studio 2026.

Microsoft bid farewell to the year-old naming method, and the new Visual Studio 2026 brings:

• • • AI Copilot deep integration **: Agent mode native support, no additional plug-ins are needed
• • • Performance has been significantly improved **: Faster startup speed and lower memory usage
• • • New UI design **: Modern interface, better dark theme support
• • • Git integration enhancements **: A more powerful code review and merge experience
• • • Full support for. NET 10 and C#14 **: out-of-the-box support for the latest technology stack

---

* * 3. Deep integration of. NET + AI **

In 2025, the integration of. NET and AI will reach a new height. Microsoft provides a complete AI development tool chain from the underlying abstraction to the upper framework.

3.1 Microsoft.Extensions.AI (MEAI)

MEAI is a unified abstraction layer for AI services launched by Microsoft and will be updated to version 10.0 in 2025.

* * Core Values **

The core question MEAI solves is: ** How to make your code independent of a specific AI provider **.

就像ILogger让你的日志代码不依赖于特定的日志框架一样，MEAI的IChatClient让你的AI代码不依赖于OpenAI、Azure或其他任何提供商。

* * Core interface **

// 聊天客户端接口
publicinterfaceIChatClient
{
    Task<ChatCompletion> CompleteAsync(
        IList<ChatMessage> chatMessages,
        ChatOptions? options = null,
        CancellationToken cancellationToken = default);
}

// 嵌入生成接口
publicinterfaceIEmbeddingGenerator<TInput, TEmbedding>
{
    Task<GeneratedEmbeddings<TEmbedding>> GenerateAsync(
        IEnumerable<TInput> values,
        EmbeddingGenerationOptions? options = null,
        CancellationToken cancellationToken = default);
}

* * Supported providers **

MEAI supports multiple AI providers:

• OpenAI / Azure OpenAI
• Anthropic Claude
• Google Gemini
• Ollama (local model)
• More communities realize...

* * Middleware model **

MEAI supports the middleware model and can insert various processing logic into the AI call chain:

IChatClient client = new ChatClientBuilder(openAIClient)
    .UseLogging()           // 日志记录
    .UseFunctionInvocation() // 函数调用
    .UseRetry()             // 重试策略
    .Build();

3.2 Semantic Kernel (SK)

Semantic Kernel is an AI orchestration framework launched by Microsoft in 2023. It will be updated to version 1.68 in 2025, making its functions more mature.

* * Agent Framework **

SK's Agent framework is its core highlight, supporting the creation of intelligent agents that can independently plan and execute tasks:

var agent = new ChatCompletionAgent
{
    Name = "CodeReviewer",
    Instructions = "你是一个代码审查专家，帮助开发者改进代码质量。",
    Kernel = kernel
};

var response = await agent.InvokeAsync("请审查这段代码...");

* * Deep integration with MEAI **

SK is now built entirely on MEAI, which means:

• You can use any MEAI-supported AI provider
• Enjoy MEAI's middleware capabilities
• Maintain code portability

3.3 Microsoft Agent Framework (MAF)

In October 2025, Microsoft officially released a preview version of the Microsoft Agent Framework, which is a new unified framework for. NET developers to build AI Agents.

* * Technical basics **

MAF did not start from scratch, but was built on Microsoft's existing AI technology stack:

• • • Semantic Kernel **: Provides powerful orchestration capabilities
• • • AutoGen **: Supports advanced multi-agent collaboration and research-driven technologies
• • • Microsoft. Extensions. AI **: Provides standardized AI building blocks

MAF is the evolution and unification of these technologies, providing. NET developers with a consistent agent development experience.

* * Core concepts **

MAF defines an agent as: ** a system that combines reasoning, context, and tools to pursue goals **.

• • • Reasoning and decision-making **: Usually driven by LLM, search algorithms, planning systems, etc. can also be used
• • • Context awareness **: External information such as dialogue history, knowledge base, and enterprise data
• • • Tool usage **: Callable capabilities such as API, MCP tools, code execution, and data query

* * Workflow type **

MAF supports multiple workflow modes to meet the needs of different scenarios:

• • • Sequential **: Agents execute in order, and results are passed along the chain
• • • Concurrent **: Multiple Agents work in parallel
• • • Handoff **: Transfer control between Agents based on context
• • • Group chats **: Agents collaborate in a shared real-time conversation space

* * Code example **

Creating an agent requires only a few lines of code:

// 创建写作Agent
AIAgent writer = new ChatClientAgent(
    chatClient,
    new ChatClientAgentOptions
    {
        Name = "Writer",
        Instructions = "Write stories that are engaging and creative."
    });

// 创建编辑Agent
AIAgent editor = new ChatClientAgent(
    chatClient,
    new ChatClientAgentOptions
    {
        Name = "Editor",
        Instructions = "Make the story more engaging, fix grammar, and enhance the plot."
    });

// 组合为工作流
Workflow workflow = AgentWorkflowBuilder.BuildSequential(writer, editor);
AIAgent workflowAgent = await workflow.AsAgentAsync();

// 执行
var response = await workflowAgent.RunAsync("Write a short story about a haunted house.");

* * DevUI for Developers **

MAF提供了直观的DevUI，帮助开发者可视化设计、调试和监控Agent和工作流，可通过安装Microsoft.Agents.AI.DevUI包快速集成。

* * Production-ready features **

• • • ASP.NET integration **: Expose agent services using the familiar Minimal API pattern
• 依赖注入：通过AddAIAgent注册，支持Keyed Services
• • • OpenTelemetry **: Built-in observability, one line of code enables telemetry
• • • Evaluation testing **: Integrate Microsoft.Extensions. AI. Evaluations for quality assessment

* * 3.4 protocol SDK fully supports **

The. NET ecosystem now fully supports major protocols in the AI space, including MCP, A2A and AG-UI protocols.

MCP C# SDK

官方的MCP .NET实现，基于 ModelContextProtocol 库，让你轻松创建MCP Server。.NET 10提供了项目模板，一行命令即可创建：

dotnet new mcpserver -n MyMcpServer

使用 [McpServerTool] 属性定义工具：

[McpServerTool]
[Description("Gets a random number between min and max.")]
public int GetRandomNumber(
    [Description("Minimum value")] int min,
    [Description("Maximum value")] int max)
{
    return Random.Shared.Next(min, max + 1);
}

[McpServerTool]
[Description("Describes random weather in the provided city.")]
public string GetCityWeather(
    [Description("Name of the city")] string city)
{
    var weather = new[] { "sunny", "rainy", "cloudy" };
    return $"The weather in {city} is {weather[Random.Shared.Next(weather.Length)]}.";
}

更棒的是，MCP Server可以直接打包为NuGet包分发，用户通过dnx命令一键安装使用。

A2A C# SDK

The A2A.NET SDK provides a complete implementation of inter-agent communication:

dotnet add package A2A
dotnet add package A2A.AspNetCore

服务端：通过 TaskManager 管理Agent，使用 MapA2A 映射端点：

var taskManager = new TaskManager();
var agent = new EchoAgent();
agent.Attach(taskManager);  // 注册回调：OnAgentCardQuery、OnMessageReceived

app.MapA2A(taskManager, "/echo");

客户端：使用 A2ACardResolver 发现Agent，A2AClient 发送消息：

var cardResolver = new A2ACardResolver(new Uri("https://localhost:5048/echo"));
var agentCard = await cardResolver.GetAgentCardAsync();

var client = new A2AClient(new Uri(agentCard.Url));
var response = await client.SendMessageAsync(new MessageSendParams
{
    Message = new AgentMessage
    {
        Role = MessageRole.User,
        Parts = [new TextPart { Text = "Hello!" }]
    }
});

MAF集成：Microsoft Agent Framework 也提供了A2A协议支持（Microsoft.Agents.A2A），可以将MAF Agent暴露为A2A服务端，实现跨框架的Agent互操作。

* *. NET support for AG-UI **

The MAF framework has built-in AG-UI support to enable streaming interaction between agents and user interfaces:

dotnet add package Microsoft.Agents.AI.Hosting.AGUI.AspNetCore  # 服务端
dotnet add package Microsoft.Agents.AI.AGUI                      # 客户端

服务端：使用 AddAGUI() 注册服务，MapAGUI() 映射端点：

builder.Services.AddAGUI();

AIAgent agent = chatClient.AsIChatClient().CreateAIAgent(
    name: "Assistant",
    instructions: "你是一个友好的AI助手。");

app.MapAGUI("/", agent);  // 自动处理SSE流式响应

客户端：使用 AGUIChatClient 连接服务端，RunStreamingAsync 接收流式响应：

AGUIChatClient chatClient = new(httpClient, "http://localhost:8888");
AIAgent agent = chatClient.CreateAIAgent(name: "client");
AgentThread thread = agent.GetNewThread();

await foreach (var update in agent.RunStreamingAsync(messages, thread))
{
    if (update.Contents.OfType<TextContent>().FirstOrDefault() is { } text)
        Console.Write(text.Text);
}

* * 3.5. NET support for Agent Skills **

Agent Skills是一种轻量级、开放的格式，用于扩展AI Agent的能力。一个Skill本质上就是一个包含SKILL.md文件的文件夹：

skill-name/
├── SKILL.md          # 必需：技能描述和使用说明
├── scripts/          # 可选：可执行脚本
│   └── tool.cs
├── references/       # 可选：详细参考文档
│   └── REFERENCE.md
└── assets/           # 可选：静态资源
    └── template.json

• • SKILL. md format **: Contains YAML frontmatter and Markdown text:

---
name: split-pdf
description: Split PDF files into separate single-page documents. Use when you need to divide a PDF into multiple files.
license: MIT
---

# Split PDF

将PDF文件拆分为多个单页文件。

## 使用方法
dotnet scripts/split-pdf.cs input.pdf output-dir/

• • Unique advantages of. NET 10 **: File-Based Apps make C#ideal for Agent Skills scripts:

#!/usr/bin/env dotnet
#:package PdfSharpCore@1.3.65
#:package Spectre.Console@0.49.1

using PdfSharpCore.Pdf;
using PdfSharpCore.Pdf.IO;
using Spectre.Console;

if (args.Length < 2)
{
    AnsiConsole.MarkupLine("[red]用法: dotnet split-pdf.cs <PDF文件> <输出目录>[/]");
    return1;
}

var pdfPath = args[0];
var outputDir = args[1];
Directory.CreateDirectory(outputDir);

usingvar doc = PdfReader.Open(pdfPath, PdfDocumentOpenMode.Import);
for (int i = 0; i < doc.PageCount; i++)
{
    usingvar output = new PdfDocument();
    output.AddPage(doc.Pages[i]);
    output.Save(Path.Combine(outputDir, $"page_{i + 1:D3}.pdf"));
}

AnsiConsole.MarkupLine($"[green]✅ 拆分完成！生成 {doc.PageCount} 个文件[/]");
return0;

与Python相比，.NET File-Based Apps的优势：依赖声明内联在文件中（#:package）、编译时类型检查、支持Native AOT编译实现毫秒级启动。

---

* * 4. Outlook 2026 **

* * 4.1 Technology Trend Prediction **

* * Agent capabilities continue to increase **

In 2026, we will see more powerful Agents:

• • • Longer context **: Handle more complex tasks
• • • Better planning **: Improved success rate for multi-step tasks
• • • Stronger tool invocation **: Support for more complex tool combinations

* * Multimodal AI becomes standard **

The understanding and generation of images, audio, and videos will become basic capabilities of AI rather than advanced functions. NET developers need to be prepared to work with multimodal data.

* * Evolution of AI native application architecture **

• "AI native"* will become a new architectural paradigm, just as "cloud native" has changed the way we build applications. We need to rethink:

• How to design an AI-friendly API
• How to build services that can be called by AI
• How to deal with the uncertainty of AI

* * Opportunities for 4.2. NET Developers **

* * Demand for AI Agent development grows **

Enterprise demand for customized AI Agents is exploding. Gartner predicts that by 2026, more than 80% of enterprises will deploy some form of AI Agent. Developers familiar with. NET and AI have the opportunity to excel in the following areas:

• • Enterprise application scenarios **:

• • • Intelligent customer service agent **: A 7x24-hour intelligent Q & A system based on enterprise knowledge base
• • • Business process automation **: Automation of repetitive tasks such as order processing, report generation, and data analysis
• • • Code Assistant **: Code review, document generation, and technical debt Detection Tools within the enterprise
• • • DevOps Agent **: Automated deployment, monitoring alarm analysis, fault diagnosis

• • Technology realization path **:

• • • Develop MCP Server **: Encapsulate enterprise internal systems (ERP, CRM, database) into MCP tools for AI to call
• • • Build a multi-agent collaboration system **: Use MAF's Workflow model to realize intelligent orchestration of complex business processes
• • • Create domain-specific Agents **: Combine the advantages of the. NET ecosystem (such as finance, manufacturing, and medical) to create industry-customized solutions

• • Career development opportunities **:

• • • AI Application Architect **: Designing enterprise-level AI application architecture
• • • Agent Development Engineer **: Be proficient in MEAI/MAF and become the core of the team
• • • AI tool developer **: Develop open source MCP Server and Agent Skills to build personal brands

* *. NET's unique advantages in AI **

NET has its unique advantages in the AI field:

• • • Performance **: For AI applications that require high performance,. NET is ideal
• • • Enterprise Foundation **: A large number of enterprise systems are built based on. NET, and AI enhancement has natural advantages
• • • Tool chain matures **: Visual Studio + Copilot is one of the best AI-assisted development experiences currently available
• • • Ecological improvement **: MEAI/SK/MAF provides a complete AI development tool chain

* * Suggestions for continuous learning **

As a. NET developer, it is recommended to focus on:

1. • • Learn Vibe Coding **: Improve the efficiency of AI-assisted development
2. • • Focus on the development of AI models **: Understand the latest AI capabilities and trends
3. • • Mastering MEAI **: This is the foundation of. NET AI development
4. • • Learn the Microsoft Agent Framework **: The core framework for agent development
5. • • Understand the MCP/AG-UI/A2A protocol **: The de facto standard for AI tool development
6. • • Practical agent development **: Accumulate experience through practical projects

---

* * 5. Summary **

2025 is a milestone year for the integration of. NET and AI.

Looking back on the year:

• • • AI enters the agent era **: From code completion to autonomous task completion, AI capabilities have undergone qualitative changes
• • • Protocol standardization promotes ecological prosperity **: MCP, A2A, AG-UI and other protocols build the foundation for interoperability
• • *. NET 10 brings important upgrades **: Direct running of. cs files, performance optimization, AI integration enhancements
• • • New Visual Studio 2026 experience **: AI Copilot is deeply integrated, greatly improving development efficiency
• • *. NET AI tool chain is becoming increasingly mature **: MEAI, SK, and MAF provide complete development capabilities

This is the best era for. NET developers. We have a mature language and platform, a complete tool chain, and an active community. Now, we have strong AI help again.

Embracing AI is not a choice, but a necessity.

The future is here, are you ready?

---

• This article was written on December 25, 2025. I wish all developers a happy New Year, good health, smooth work, all their wishes come true, everything goes well, and their families are happy! *