무슨 일이 일어났나요?
GitHub’s August 28 weekly release roundup describes updates to Copilot across Slack, Microsoft Teams, the Copilot app, CLI, JetBrains, VS Code and Visual Studio. The changes focus on shared agent work, customization, execution controls, session recovery, model selection and usage visibility.
GitHub says Copilot can now turn conversations in Slack and Microsoft Teams into shared agent sessions. By mentioning @GitHub, a team can ask Copilot to investigate problems, plan work and make changes that other participants can follow and guide. The source frames this as collaborative, visible agent work rather than a private exchange between one developer and the assistant. It does not explain which actions Copilot can take in a given workspace, what approval steps are required before changes are made, or whether the Slack and Teams capabilities are available to every Copilot customer.
The Copilot app receives several changes. GitHub says its Customize tab is now generally available and brings MCP servers, plugins, skills and canvases together in one place. The same tab can turn Azure DevOps issues and pull requests into Copilot sessions. The app also gains experimental support for working in a Linux environment through Windows Subsystem for Linux, the ability to split and move any tab, and an option to send a browser preview to an external browser from a tab’s context menu. These descriptions indicate a broader configuration surface, but the source does not define the supported plugins, skills or MCP servers, nor does it state what restrictions apply to the experimental Linux support.
The CLI changes address both configuration and continuity. GitHub says users can set preferred execution and permission modes for new sessions through defaultMode and defaultPermissionMode. New experiences in /plugin, /mcp and /skills are intended to make those components easier to manage. Sessions that did not exit cleanly can be restored, including sessions interrupted in the middle of a turn. GitHub also says Copilot CLI now runs on a native Rust runtime while its terminal interface remains built in TypeScript, which it presents as a significant performance improvement. No , workload, latency figure or compatibility information is included in the source.
JetBrains users receive enterprise controls for plugins, MCP servers, telemetry and agent permission modes. In VS Code 1.135, GitHub says users can continue recent Copilot or Claude agent sessions from other applications, request a second opinion from a complementary model, use a single-pane Agents layout and find simpler session controls and session details. VS Code also adds detailed chat usage by model for each chat turn. The source points readers to full VS Code release notes, but those notes are not included here, so the roundup does not establish the complete behavior or availability of each change.
왜 중요한가요?
The release moves Copilot beyond an individual coding assistant toward a more configurable system that can participate in team conversations, connect to tools and preserve work across development environments. The source describes capabilities and availability, but provides no independent testing, adoption data, performance measurements or evidence that the changes improve software quality.
The most consequential theme is control. Copilot is being presented as a system whose behavior can be shaped through execution modes, permission modes, plugins, MCP servers and skills. That matters because coding assistants increasingly interact with repositories, development tools and external services. Centralizing those settings may make them easier for teams to discover and manage, while also creating a larger administrative surface that organizations must understand before enabling it.
The Slack and Teams additions make the assistant a shared participant in software work. A team can apparently watch an investigation or plan unfold and guide it together, which could reduce the gap between a developer’s private session and a group’s operational discussion. At the same time, a shared session raises practical questions about authorization, ownership, audit trails and responsibility for changes. GitHub’s source says that teammates can follow and guide the work, but does not say how disagreements, sensitive repositories or unauthorized requests are handled.
The app’s Customize tab could be useful for organizations that use several extension mechanisms. MCP servers, plugins and skills can affect what Copilot can access or do, while canvases and Azure DevOps connections expand the kinds of work that can be brought into a session. Bringing these elements together may simplify setup, but the source gives no security review, isolation guarantees or explanation of how permissions are inherited. Readers should therefore treat the description as a product claim, not as evidence that every configuration is safe or interoperable.
The CLI’s session restoration and model-specific usage details address two persistent operational needs: recovering work after interruption and understanding how an AI service is being consumed. Default permission settings may help standardize behavior across users, and per-turn model usage could help teams monitor premium-model costs. Yet the source does not provide a cost model, retention policy, recovery guarantee or evidence that Rust changes performance for typical workloads. It also does not say whether restored sessions preserve every state element or require users to review actions before continuing.
대화형 메커니즘: 실제로 작동하는 방식
이 개발의 이면에 있는 기본 기술을 대화식으로 살펴보세요.
An agent must create a draft calendar event for Tuesday at 2 p.m. Which evidence would establish the requested result?
다음에 무엇을 볼 것인가
The important follow-up is whether these features work reliably in real team workflows and how organizations govern permissions, plugins, MCP servers, telemetry and premium-model consumption. GitHub’s roundup does not specify the rollout scope for every , the exact limits of shared agent sessions, or how much faster the native Rust runtime is.
The first question is rollout and scope. GitHub labels the Customize tab generally available, but labels Linux support through WSL experimental. The roundup does not give a complete availability matrix for Slack, Teams, JetBrains, VS Code or Visual Studio, and it does not state whether access varies by Copilot plan, organization policy, operating system or editor version. Follow-up documentation will be needed to establish who can use each capability and under what limits.
Security and governance deserve close attention as shared sessions and configurable extensions reach more users. Teams will need clear answers about which identities Copilot uses, how permission modes constrain actions, what MCP servers can access, how plugin and skill changes are reviewed, and what telemetry administrators can see. The source announces enterprise controls for JetBrains and usage visibility in VS Code and Visual Studio, but it does not describe audit logs, approval workflows, data retention or incident-response mechanisms.
Reliability is another open issue. Session restoration could reduce lost work after an interrupted turn, but the source does not quantify restoration success or explain how partially completed actions are represented. Similarly, the native Rust runtime is described as significantly faster without supporting measurements. Independent tests should compare startup time, command latency, resource use and behavior across representative repositories, terminals and operating systems.
The Visual Studio portion overlaps with GitHub’s separate August update covering favorite models, model comparisons, reasoning-effort controls, shared custom agents, plan-consumption views and Git-agent reviews of uncommitted changes or commits. The weekly roundup adds context by placing those changes alongside updates in other Copilot surfaces. The broader practical test will be whether these controls produce more predictable workflows for teams, and whether the added flexibility makes Copilot easier to govern or simply more complex to configure.