{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreicltsvbwhtzvsg6nnbjuuyv2lecxqjxl47ukhxcxnmyh6cz4ogosu",
    "uri": "at://did:plc:25rdn5elo5izoxrmtis34zuk/app.bsky.feed.post/3mpome26iq572"
  },
  "coverImage": {
    "$type": "blob",
    "ref": {
      "$link": "bafkreib6k6esexfa7irgixxvgn5d7u7jjohoz7of37gxjf2op44zrnbdwu"
    },
    "mimeType": "image/webp",
    "size": 307740
  },
  "path": "/restuananda/microsofts-99-ai-skills-are-interesting-the-bigger-story-is-what-they-represent-pl3",
  "publishedAt": "2026-07-02T17:36:46.000Z",
  "site": "https://dev.to",
  "tags": [
    "ai",
    "dotnet",
    "microsoft",
    "softwareengineering",
    "https://fastwork.id/byob/7pFhYwWqZd?openExternalBrowser=1&source=byob",
    "https://www.upwork.com/services/product/development-it-custom-business-website-development-2067932090568948507?ref=project_share"
  ],
  "textContent": "##  AI coding assistants already understand .NET. The next challenge is teaching them how _your_ team builds software.\n\nOne of the most interesting things happening in software engineering right now isn't the arrival of another language model.\n\nIt's the gradual realization that intelligence alone isn't enough.\n\nOver the past couple of years, AI coding assistants have become remarkably capable. Whether you're using GitHub Copilot, Claude Code, Cursor, Codex, or another modern coding assistant, it's difficult not to be impressed by how quickly they can generate APIs, write unit tests, explain unfamiliar code, or scaffold entire applications.\n\nIn many cases, they already understand programming languages better than most developers ever will.\n\nAsk them about C#, ASP.NET Core, Entity Framework, LINQ, Minimal APIs, dependency injection, or asynchronous programming, and the answers are usually accurate, well-structured, and surprisingly practical.\n\nYet despite all of that progress, I continue hearing the same complaint from engineering teams.\n\n_\"The AI still doesn't understand our project.\"_\n\nAt first, that sounds like a limitation of the model itself.\n\nThe more I think about it, the more I believe it's actually a limitation of context.\n\n##  Understanding a Language Isn't the Same as Understanding a Codebase\n\nKnowing how .NET works is only one small part of software engineering.\n\nEvery engineering organization develops its own way of building software.\n\nSome teams organize their applications around Clean Architecture.\n\nOthers prefer Vertical Slice Architecture.\n\nSome rely heavily on CQRS with MediatR.\n\nOthers use repository patterns, Result, domain events, feature folders, or entirely custom conventions that have evolved over years of development.\n\nThen there are the business rules.\n\nThe naming conventions.\n\nThe deployment pipeline.\n\nThe testing philosophy.\n\nThe internal libraries.\n\nThe security requirements.\n\nThe architectural decisions that exist nowhere except inside the heads of experienced engineers.\n\nThis is the context that makes one company's codebase fundamentally different from another's.\n\nModern language models understand C#.\n\nWhat they don't automatically understand is _your organization_.\n\nEvery new conversation begins from almost zero.\n\nThe assistant has to rediscover your architecture, infer your conventions, and guess how your team prefers to solve problems.\n\nThat's an incredibly expensive way to collaborate.\n\n##  Why Microsoft's Repository Matters\n\nThis is why Microsoft's recently open-sourced **dotnet/skills** repository caught my attention.\n\nAt first glance, the announcement appears relatively straightforward.\n\nMicrosoft published nearly one hundred reusable AI skills covering common .NET development tasks such as ASP.NET Core, Entity Framework, testing, project modernization, upgrades, and AI application development.\n\nFor many developers, the headline became the number.\n\nNinety-nine skills.\n\nAn impressive collection.\n\nBut I think the number is the least interesting part.\n\nThe real innovation is the idea behind it.\n\nInstead of repeatedly explaining the same workflows every time an AI assistant starts a new session, those instructions can be packaged into reusable skills that compatible AI agents automatically load whenever they're relevant.\n\nThe assistant doesn't need to be reminded how to approach a particular .NET task.\n\nThe knowledge already exists.\n\nThe context is reusable.\n\n##  Microsoft Can Teach AI About .NET\n\nOnly You Can Teach AI About Your Company\n\nThis is where I think the conversation becomes much more interesting.\n\nMicrosoft can teach AI how to write better .NET applications.\n\nOnly your engineering team can teach AI how to build software _your way._\n\nImagine every architectural principle your team follows becoming reusable knowledge.\n\nYour preferred folder structure.\n\nYour API design guidelines.\n\nYour naming conventions.\n\nHow authentication works.\n\nHow repositories are organized.\n\nHow pull requests should be reviewed.\n\nHow services communicate.\n\nHow logging is implemented.\n\nHow feature flags are introduced.\n\nHow migrations are managed.\n\nHow domain models evolve.\n\nInstead of explaining these ideas repeatedly, they become part of your organization's shared engineering memory.\n\nEvery AI-assisted development session starts with context rather than assumptions.\n\nThat changes the relationship between engineers and AI completely.\n\n##  From Prompts to Organizational Knowledge\n\nOne pattern has become increasingly obvious over the last year.\n\nMost developers spend far too much time repeating themselves.\n\nEvery prompt includes reminders about coding standards.\n\nEvery conversation explains the same architecture.\n\nEvery new feature starts with another description of how the application is structured.\n\nEventually, prompt engineering becomes organizational overhead.\n\nReusable skills solve a different problem.\n\nInstead of writing better prompts, teams begin capturing institutional knowledge.\n\nKnowledge stops living inside Slack conversations, onboarding documents, or the minds of senior engineers.\n\nIt becomes something AI can actively use while generating software.\n\nThat may sound like a subtle difference.\n\nI don't think it is.\n\nIt's the difference between asking someone to remember instructions every day and giving them a well-designed handbook they can reference automatically.\n\n##  The Competitive Advantage Is Shifting\n\nWhen AI coding assistants first became popular, much of the conversation focused on choosing the best model.\n\nShould you use Claude?\n\nCopilot?\n\nCursor?\n\nCodex?\n\nThe assumption was that better models would naturally create better engineering outcomes.\n\nI'm no longer convinced that's where the biggest advantage will come from.\n\nAs foundation models continue improving, the performance gap between them will likely become smaller.\n\nWhat becomes difficult to copy isn't the model.\n\nIt's the organizational knowledge surrounding it.\n\nEvery engineering team has accumulated years of architectural decisions, operational experience, coding conventions, business rules, deployment strategies, and technical trade-offs.\n\nThat knowledge is incredibly valuable.\n\nUntil recently, most of it existed only inside documentation or experienced engineers.\n\nNow it can become something AI actively participates in.\n\nThat's a much more durable competitive advantage than simply paying for access to the latest language model.\n\n##  Software Engineering Is Becoming More About Context\n\nThe longer I work with AI-assisted development, the more I believe that software engineering is quietly shifting from writing code toward managing context.\n\nGenerating code is becoming cheaper every month.\n\nProviding the right context is becoming increasingly valuable.\n\nThe organizations that succeed won't necessarily be the ones with the most advanced AI models.\n\nThey'll be the ones that systematically teach those models how they build software, why they make certain architectural decisions, and what quality means inside their engineering culture.\n\nThat's knowledge no foundation model can learn on its own.\n\n##  Conclusion\n\nMicrosoft's ninety-nine AI skills are undoubtedly useful.\n\nThey'll help developers work more effectively with .NET, automate common tasks, and reduce repetitive instructions during AI-assisted development.\n\nBut I think the repository points toward something much bigger.\n\nThe future isn't simply AI that understands programming languages.\n\nIt's AI that understands engineering organizations.\n\nBecause once an assistant understands not only how to write C#, but also how _your team_ designs systems, reviews code, structures projects, and makes architectural decisions, something fundamental changes.\n\nThe assistant stops feeling like a code generator.\n\nIt starts feeling like a new engineer who's already completed onboarding.\n\nAnd that may become one of the most valuable productivity improvements software engineering has seen in years.\n\n##  Building AI-Powered Software\n\nI spend much of my time helping founders, startups, and businesses build modern software systems, integrate AI into existing products, and design cloud-native architectures that can scale as businesses grow.\n\nIf you're exploring AI-assisted development, custom business software, intelligent automation, or scalable web applications, I'd be happy to help turn those ideas into production-ready solutions.\n\n**Fastwork:** https://fastwork.id/byob/7pFhYwWqZd?openExternalBrowser=1&source=byob\n\n**Upwork:** https://www.upwork.com/services/product/development-it-custom-business-website-development-2067932090568948507?ref=project_share\n\n##  About Me\n\nMost days you'll find me building software, experimenting with AI, exploring distributed systems, and writing about the ideas I encounter along the way.\n\nI'm particularly interested in software engineering, artificial intelligence, cloud computing, system architecture, and how emerging technologies are changing the way we design, build, and maintain software. Through these articles, I share lessons from real projects, technical research, and observations about where our industry is heading.\n\nI don't write because I think I have all the answers. I write because some of the best ideas emerge when they're shared, challenged, and refined together.",
  "title": "Microsoft's 99 AI Skills Are Interesting. The Bigger Story Is What They Represent."
}