News
AI Subagents 'Coming Soon' to Visual Studio Copilot
Mads Kristensen, principal product manager for Visual Studio, said subagents are coming to Copilot in Visual Studio, signaling another agentic AI feature moving toward Microsoft's flagship Windows IDE after appearing first in Visual Studio Code.
"Sub agents are coming to Copilot in Visual Studio. Such as smooth implementation too. Coming soon..." Kristensen wrote in a May 6 social media post, which included a screenshot showing a larger Copilot task being delegated to a focused subtask labeled "General-purpose -- Audit EditorConfig.json schema." This image accompanied the post:
While it's impossible to tell, Kristensen's image might well have come from VS Code.
The post comes as Microsoft has been adding agentic AI functionality across both VS Code and Visual Studio, though the release cadence has often favored VS Code first. VS Code's extension-driven model has made it an early venue for Copilot agent features, while Visual Studio has been picking up many of the same concepts through updates to Copilot Chat, agent mode, cloud agents, built-in agents and custom agents.
But Kristensen has been busy adding AI functionality to the IDE, perhaps picking up the pace when he got an earful from developers about the gap between VS Code and Visual Studio in response to another post wherein he asked: "What features or extensions make you jump from Visual Studio to other IDEs and editors to perform certain tasks?" For more on that, see "Devs Diss Visual Studio's AI."
While Kristensen's post was published just minutes before this article was written -- so dev feedback is scant at this point -- at least one early commenter was enthusiastic: "Excellent. Between the last Insiders update and this on the horizon, I see you're attempting to end my reliance on Cursor. Keep it going."
What Subagents Do
In VS Code, Microsoft describes a subagent as "an independent AI agent that performs focused work, such as researching a topic, analyzing code, or reviewing changes, and reports the results back to the main agent."
The idea is to keep the main chat or main agent focused while a secondary agent handles a narrower task in its own context. VS Code's documentation lists scenarios such as researching implementation approaches before coding, running parallel code analysis, exploring multiple possible solutions, using specialized review agents for security or performance, and running multiple models against the same problem.
When a subagent runs in VS Code, it appears in chat as a collapsible tool call. The default collapsed view can show the custom agent name and the currently running tool, such as reading a file or searching the codebase. Expanding the item exposes details including tool calls, the prompt passed to the subagent and the result returned.
That matches the general UI concept shown in Kristensen's image: a broader Copilot task contains a delegated subtask, with the subtask shown as its own collapsible line item.
VS Code Got Subagents First
VS Code has already been living with this idea for a while. Around GitHub Universe 2025, GitHub described context-isolated sub-agents in VS Code Insiders as a way to let Copilot hand off focused work -- such as research or test-driven development tasks -- without filling up the main chat context.
That is the basic appeal of subagents: they give the main agent somewhere else to send a side quest. Instead of one Copilot chat trying to read files, search the codebase, compare approaches and implement changes all in the same context window, a subagent can go off, do a narrower job and return a summary.
Microsoft explained the same concept in its VS Code post on "A Unified Experience for All Coding Agents," describing a runSubagent tool that lets an agent spin up a separate worker with its own context. Only the result comes back to the main chat, which is the point: less clutter in the main conversation and less risk that background research consumes the space needed for the actual coding task.
The same shift was also covered in Visual Studio Magazine's "Microsoft Details How 'Agents Took Over VS Code in 2025'," which examined VS Code's move toward a unified agent experience, custom agents and related orchestration features.
From there, subagents started looking less like a one-off tool and more like part of VS Code's broader agent orchestration model. The VS Code 1.107 release notes described agents running other agents as subagents, with the added twist that custom agents could be used for the delegated work. That means the main agent could hand a task to a more specialized agent -- for example, one configured for research, review or another focused workflow.
That November 2025 update was covered in a Visual Studio Magazine article, VS Code 1.107 article, which noted the continued expansion of multi-agent orchestration and model-management capabilities in the editor.
The feature kept expanding in VS Code 1.109, where Microsoft added support for subagents running in parallel and improved the UI so developers could see more of what each one was doing. That release also introduced an experimental search subagent, meant to handle the kind of iterative codebase searching that can otherwise take over the main context window.
Visual Studio Magazine covered that release in "VS Code 1.109 Deemed a Multi-Agent Development Platform," and followed with "Hands On with New Multi-Agent Orchestration in VS Code," which looked at the workflow from a developer's perspective.
Custom, Parallel and Model-Specific Subagents
VS Code's current docs describe subagents as more than a single command. They are now part of a broader agent orchestration model.
Subagents are typically agent-initiated, not directly invoked by users in chat. The main agent decides when context isolation is useful, starts the subagent, passes only the relevant subtask, receives a summary and continues. Users can hint at subagent use by asking for isolated research or parallel analysis, but the system is designed so the main agent can decide when to delegate.
VS Code also supports custom agents as subagents. Custom agents can specify their own tools, instructions and models, overriding the defaults inherited from the main session. This allows workflows such as a research agent with read-only tools, a security agent focused on vulnerability review, a performance agent focused on bottlenecks, or a planning agent used before implementation.
The docs also describe model selection for subagents. The model can be chosen explicitly by the main agent when invoking the runSubagent tool, through the custom agent's model property in its .agent.md frontmatter, or by falling back to the parent conversation's model.
Nested subagents are disabled by default, according to the docs, to avoid recursion problems. VS Code includes a setting, chat.subagents.allowInvocationsFromSubagents, that can allow recursive delegation up to a maximum depth of 5.
Microsoft summarized the broader direction in a February 2026 VS Code blog post titled "Your Home for Multi-Agent Development." That post said subagents are context-isolated agents that run independently from the main session, with only the final result flowing back. It also said VS Code could now run multiple subagents in parallel and show more detail about what they are doing.
Where Visual Studio Stands
Visual Studio already has several agentic AI features, but Microsoft's public Visual Studio documentation does not yet document subagent orchestration as an available Visual Studio feature.
The Visual Studio Learn page for GitHub Copilot agent mode says Copilot in Visual Studio can take a high-level task, create a plan, make code edits, run terminal commands, invoke tools and iterate on outcomes such as build results and unit-test failures. The page says agent mode can use built-in tools, Model Context Protocol (MCP) tools and agent skills, and it points to built-in agents such as @debug, @profiler, @test and @vs.
Another Visual Studio Learn page on built-in and custom agents says Visual Studio includes curated built-in agents for debugging, profiling, testing and modernization, along with custom agents tailored to team workflows. It describes access through an agent picker or @ syntax, but not a parent agent delegating work to subagents.
Microsoft also discussed custom Visual Studio agents in a February 2026 Visual Studio Blog post, "Custom Agents in Visual Studio: Built in and Build-Your-Own agents." That post describes built-in and custom agents, but again does not present subagents as a Visual Studio feature. The April 2026 Visual Studio update added more agentic infrastructure, including cloud agent integration, but that is a separate workflow in which work is handed off to a cloud agent that can open a pull request.
Kristensen's post therefore points to another catch-up step for Visual Studio: not just having named agents or cloud agents, but allowing Copilot to break work into delegated sub-agent tasks inside the IDE.
The Visual Studio Angle
For Visual Studio developers, the practical value is context management and task decomposition. A single agent conversation can get crowded as it reads files, explores implementation patterns, checks docs, reviews tests and reasons through alternatives. Subagents separate those side investigations from the main thread.
In a large .NET solution, that could mean one subagent audits configuration files, another reviews test coverage, another searches for prior implementation patterns and another checks security-sensitive code paths. The main agent then receives summaries rather than every intermediate tool call and file-reading step.
That does not eliminate developer review. Subagents still sit inside Copilot's agent workflow, where developers need to inspect proposed changes, review tool use and decide what to accept. But the model changes from one assistant doing everything in one thread to a coordinator agent that can delegate focused work to narrower agents and then synthesize the results.
That distinction matters because Visual Studio's AI roadmap has increasingly moved from autocomplete and chat toward agentic workflows: plan the work, inspect the project, use tools, edit code, run commands, test results and iterate. Subagents add another layer to that model by letting the AI system split the work internally.
For now, VS Code remains the documented reference point for how Copilot subagents work in Microsoft's developer tooling. Kristensen's post indicates Visual Studio is next.
About the Author
David Ramel is an editor and writer at Converge 360.