Skip to content

VS Code 1.104: Smarter AI Model Selection & Enhanced Security

  • News
VS Code 1.104: Smarter AI Model Selection & Enhanced Security

Exploring VS Code 1.104: AI Model Selection and Agent Security Up Close

John: Hey everyone, welcome back to the blog! I’m John, your go-to guy for breaking down the latest in AI and tech. Today, we’re diving into Visual Studio Code’s version 1.104 update, which is all about smarter AI integration—specifically, emphasizing AI model selection and beefing up agent security. This release, hot off the presses as of early September 2025, is making waves for how it streamlines coding with AI while keeping things safe. If you’re a beginner or intermediate coder, stick around; we’ll make this approachable.

Lila: Hi John! I’m Lila, and as someone just getting into coding, VS Code is my daily driver. But AI model selection and agent security sound a bit advanced. Can you start with the basics? What’s this update really about?

The Basics of VS Code 1.104

John: Absolutely, Lila. Visual Studio Code, or VS Code for short, is Microsoft’s free, open-source code editor that’s super popular among developers. The 1.104 update, released just a couple of days ago on September 11, 2025, focuses on enhancing GitHub Copilot, its AI-powered assistant. The big highlights are automatic model selection in Copilot Chat and improved security for AI agents. According to the official GitHub Changelog, this means Copilot can now smartly pick the best AI model for your query, switching between options like those from OpenAI, Anthropic, or Google to give you faster, more accurate responses.

John: Oh, and if you’re into automating workflows alongside your coding—maybe integrating VS Code with other tools—our deep-dive on Make.com covers features, pricing, and use cases in plain English—worth a look: Make.com (formerly Integromat) — Features, Pricing, Reviews, Use Cases.

Lila: That sounds helpful! So, auto model selection—does that mean I don’t have to manually choose which AI brain to use?

John: Spot on. Before this, you might have stuck with one model, but now Copilot evaluates your chat query and picks the optimal one automatically. For example, if you’re asking for code optimization, it might tap into a model that’s great at efficiency, leading to better productivity. Neowin reported that this intelligent switching is a game-changer for workflows, making AI feel more intuitive.

Key Features in Action

Lila: Okay, got it. Now, what about these AI agents? I’ve heard the term but don’t fully understand. How does security come into play here?

John: Great question. AI agents in VS Code, especially in “agent mode,” are like smart helpers that can autonomously edit files in your workspace. Think of them as a robot sidekick that not only suggests code but actually makes changes for you. The 1.104 update previews enhancements to this, including safeguards to prevent mishaps. For instance, agents can now operate with better boundaries to avoid accidentally messing with critical files.

John: From what I’ve seen on verified sources like InfoWorld, this update addresses potential risks head-on. Here’s a quick list of standout features:

  • Auto Model Selection: Dynamically chooses the best AI model for chats, improving response quality and speed.
  • Agent Mode Security: New prompts and controls to warn users about file edits, reducing the chance of malicious or accidental changes.
  • Productivity Boosts: Faster edits in agent mode and integration with extensions for custom models.
  • Accessibility Improvements: Better chat interfaces for all users, as noted in the Windows Report article.

Lila: That list helps a lot. Like, agent mode sounds powerful, but scary if it can delete files. How does the security actually work?

John: It’s like having a safety net. The update includes warnings about potential side effects, such as modifying config files that could affect your machine. Microsoft emphasizes user consent—agents won’t act without your okay. Plus, there’s ongoing talk in dev communities about integrating this with secure practices, drawing from real-world feedback on platforms like GitHub.

Current Developments and Trends

Lila: Are there any recent trends or examples from the web that show this in real life? I love hearing about what’s buzzing right now.

John: Definitely! As of today, September 13, 2025, searches on X (formerly Twitter) and news sites are lighting up with devs praising the auto model selection for making Copilot feel “smarter.” One verified account from a GitHub engineer tweeted about how it reduces context switching, saving time on complex projects. On the security side, there’s buzz around a new CVE—CVE-2025-55319—highlighted in forums like Windows Forum, which points out vulnerabilities in agentic AI that could lead to remote code execution if not handled properly. Microsoft is addressing this with the 1.104 patches, urging users to update immediately.

John: Real-time updates from How-To Geek note that this version, codenamed “Eclipse,” is rolling out agent mode more widely, building on previews from earlier in 2025. It’s exciting to see how it’s evolving from basic chat to full-on collaborative agents.

Challenges and How to Overcome Them

Lila: Challenges? Like, what could go wrong with all this AI power?

John: Fair point. One big challenge is security risks—agents editing files could, in theory, be exploited if there’s a vulnerability, as seen in that CVE. Another is over-reliance on AI, where beginners might skip learning fundamentals. To overcome this, start small: Use agent mode on test projects, always review changes, and pair it with learning resources. Reputable sites like SD Times have covered similar agent modes in tools like Gemini Code Assist, stressing the need for user vigilance.

Lila: Makes sense. Any tips for someone like me to stay safe?

John: Keep your VS Code updated, enable security prompts, and consider tools that monitor AI interactions. Research from AIMultiple highlights AI agent security tools that can add extra layers of protection.

Future Potential and Applications

Lila: Looking ahead, where do you see this going? Could it change how we code?

John: Oh, the potential is huge! Imagine agents handling entire workflows, from debugging to deployment, integrated with broader AI ecosystems. With updates like this, VS Code is positioning itself as a hub for AI-driven development. In the later stages of projects, when you’re documenting or presenting code, tools like Gamma can supercharge that process. If creating documents or slides feels overwhelming, this step-by-step guide to Gamma shows how you can generate presentations, documents, and even websites in just minutes: Gamma — Create Presentations, Documents & Websites in Minutes.

John: Trending discussions on X suggest we’ll see more integrations with models from providers like Anthropic and Google, making VS Code even more versatile by 2026.

FAQs: Your Burning Questions Answered

Lila: Before we wrap up, can we do some quick FAQs? Like, is this update free?

John: Yes, VS Code is free, and so is this update—just download from the official site. Another common one: Does auto model selection require a subscription? It works with GitHub Copilot, which has a Pro tier for advanced features, but basics are accessible.

Lila: And for security, should I be worried about that CVE?

John: Not if you update promptly—Microsoft’s response is solid, as per their security center.

John: As a quick CTA, if you’re automating around VS Code, revisit our Make.com guide for seamless integrations: Make.com (formerly Integromat) — Features, Pricing, Reviews, Use Cases.

John: Reflecting on this, VS Code 1.104 is a smart step forward, balancing innovation with safety—it’s empowering coders without overwhelming them. The focus on AI selection and agents shows Microsoft’s commitment to practical, secure tools. Can’t wait to see how devs build on this.

Lila: My takeaway? This update makes AI in coding less intimidating and more helpful—I’ll definitely try agent mode on a small project soon!

This article was created based on publicly available, verified sources. References:

Tags:

Leave a Reply

Your email address will not be published. Required fields are marked *