According to Neowin, Microsoft shared its Visual Studio AI roadmap as of November 2025, outlining work the company is doing and plans to do with AI-powered agentic experiences. The roadmap focuses on smarter, faster developer tools but Microsoft explicitly stated these items aren’t commitments or guarantees for delivery. Key investigations include user-created custom agents, test agents, debugger agents, and running multiple Visual Studio Agents concurrently. The company is also working on implementing the full MCP specification for secure development stack integration and wants to expand access to evaluate the latest models including GPT 5 Codex in Chat. Microsoft emphasized these improvements will launch over coming months, if they materialize at all.
Sponsored content — provided for informational and promotional purposes.
The agent explosion is coming
Here’s the thing about Microsoft‘s vision – they’re betting big on specialized AI agents. We’re talking about dedicated test agents, debugger agents, and even letting users create their own custom agents. But can developers actually manage all these different AI personalities? Running multiple agents concurrently sounds powerful, but it also risks creating chaos in the development environment. I’ve seen how complex development workflows can get – adding multiple AI agents into the mix could either be revolutionary or a complete mess.
The MCP play is smarter than it looks
Microsoft’s push to implement the full MCP specification is actually pretty strategic. This isn’t just about better tool integration – it’s about locking developers into the Microsoft ecosystem. When you can securely integrate your entire development stack, why would you leave? The company’s focus on organization allowlists for MCP servers shows they’re thinking about enterprise concerns from the start.
Auto model selection could be a game-changer
The automatic model routing feature is probably the most practical improvement on the list. Developers shouldn’t have to manually switch between models for different tasks – that’s just busywork. Automatically routing prompts to the most appropriate model could actually save real time and mental energy. And let’s be honest – with GPT 5 Codex potentially coming to Chat, the quality improvements could be significant. But I wonder how transparent this routing will be – will developers understand why certain models are chosen?
But here’s the catch
Microsoft’s constant reminders that these aren’t commitments tell you everything. They’re throwing ideas at the wall to see what sticks. The investigation into running multiple agents and slash commands for prompts sound great, but will they actually ship? This feels like Microsoft trying to keep pace with the rapid AI innovation happening elsewhere while managing expectations. Basically, don’t rearrange your development workflow around any of this just yet.
