Microsoft released Visual Studio Code 1.145 on November 4, and the update turns the world’s most popular code editor into a meeting place for rival AI coding agents. Nested sessions can now run different agent harnesses inside one parent session, so a Copilot chat can ask Claude to review its plan while Codex tests the same code beside it. The release also brings remote app previews, five-field cron schedules for local automations, quieter chat views, and tighter enterprise controls over MCP servers.
One Session, Three Agents
The headline change is mixed harness sessions, now in preview. Previously, every nested session inherited its parent’s harness, so getting a second opinion from another agent meant starting a fresh session and losing the grouping. Now a developer can keep related conversations together while each nested chat runs its own harness, model, agent, and permission levels.
The workflow is concrete: ask Claude to review a plan that a Copilot chat wrote, or hand Codex the same task, side by side in the same session. A harness picker appears for new nested sessions, and chats keep their history and harness when moved. The feature needs the chat.agentHost.multipleChatProviders.enabled setting, and an organization can disable providers through managed settings.
Remote Sessions Feel Local
Remote agent sessions feel more local in this release. When an agent runs on a remote host, the integrated browser in the Agents window now loads pages through that host, so a developer can preview an app the agent just started in place. Remote canvases work the same way. Both are in preview and need workbench.browser.enableRemoteProxy; canvases also need chat.canvases.enabled.
Cron Schedules and Quieter Chats
Recurring agent work gets a proper scheduler. Local automations now accept five-field cron expressions, so a team can run a code review every weekday at nine in the morning instead of using presets. The dialog validates the expression and describes the resulting schedule. Cron stays local-only, and workspace custom agents can now be selected when creating or editing an automation.
Long chats get easier to read. A new verbosity setting offers three view modes: Conversation groups reasoning and tool calls into collapsed sections, Verbose expands every step, and Compact, the default, previews tool calls while they run and then collapses them into summary rows. Switching modes changes only the display, not the session’s context or the work the agent performs.
The Local Harness Loses Three Tools
The release also takes tools away from the Local harness. The Install Extension, Run VS Code Command, and Create New Workspace tools are gone, along with the /new slash command and the Generate New Workspace link on the Welcome page.
Microsoft’s replacement is deliberately manual: agents now list needed extensions in .vscode/extensions.json for the developer to review, and a project starts by opening an empty folder and describing it to an agent. Teams should audit custom agents and prompt files that reference the removed tools before rolling out the update.
Our Read
Mixed harnesses are a bet on neutrality. Cursor and Windsurf won by owning the whole agent loop, but that lock-in is their weakness: when the best model changes, users must relearn a tool. VS Code already won the editor war on familiarity, so sharing one session between rival agents turns openness into the feature. The editor becomes coding’s Switzerland.
The rest of the release points the same way: agents are becoming infrastructure, not a demo. Remote previews and cron schedules treat agent sessions as long-running work that survives across machines and mornings, while the Local harness trims are a quiet admission that agents should advise more than act unsupervised.
The trade is honest: more control for developers, less autonomy for agents. It matches what teams asked for after a year of agents running loose in production code, and it puts the permission question with the developer, not the model. Separate permission levels per harness are the detail that makes the whole multi-agent session safe to use.
Why It Matters
VS Code is where most of the world’s code gets written, so a change to how its agents cooperate ripples everywhere. If rival agents can share one session here, the AI-native editors lose their main selling point, and the competition shifts back to whose model writes the best code.
For developers, the practical win is smaller but real: a second opinion from Claude or Codex without leaving the editor, the session, or the permissions they already trust. Remote previews and cron schedules add the same kind of quiet leverage, turning the editor from a place where code is typed into a place where agents keep working.
Background
VS Code ships on a monthly cadence, and 1.145 follows a year when coding agents moved from autocomplete to full collaborators. GitHub rebuilt Git so thousands of agents can push to one repository at once, Copilot expanded into Xcode and JetBrains, and Claude Code overtook Copilot in JetBrains’ 2026 developer survey. Version 1.145.1 and 1.145.2 are already out.
Source: VS Code 1.145 release notes and Neowin’s report
0 Comments