VS Code Shifts to Weekly Releases and Introduces Autopilot Mode as AI Agents Reshape Editor Development
Microsoft's Visual Studio Code moves to weekly stable releases starting with version 1.111, adding an Autopilot mode that lets AI agents operate autonomously, while Google simultaneously launches a similar auto-approve feature in Gemini Code Assist with stark security warnings.
Overview
Visual Studio Code version 1.111, released on March 9, 2026, marks the beginning of a new weekly stable release cadence for the world’s most widely used code editor. The update also introduces Autopilot, a permission mode that allows AI agents to operate autonomously — approving their own tool calls, retrying errors, and iterating without human intervention until a task is complete. The same week, Google shipped a parallel auto-approve feature in Gemini Code Assist, accompanied by documentation that describes the capability as “extremely dangerous.”
What Changed in VS Code
After a decade of monthly releases, Microsoft is moving VS Code to weekly stable builds. Distinguished engineer Kai Maetzel announced that the team will “ship a new Stable release every week,” with the previous “Endgame” freeze-and-test week folded into ongoing weekly activities. The shift was enabled in part by AI-assisted improvements to VS Code’s own development process, including one-click test plan generation from feature request issues and automated verification steps.
The headline feature of 1.111 is Autopilot, a new tier in VS Code’s agent permission system. Where the existing “Bypass Approvals” mode already let agents skip individual confirmation prompts, Autopilot goes further: the agent continues working autonomously until it calls a task_complete tool to signal it is done. It auto-approves all tool calls, automatically retries on errors, and auto-responds to tool prompts that would otherwise stall the session.
VS Code’s release documentation includes a direct warning: “Bypass Approvals and Autopilot bypass manual approval prompts and ignore your configured approval settings, including for potentially destructive actions like file edits, terminal commands, and external tool calls.” Users must confirm they understand the security implications before enabling the mode. In the stable channel, Autopilot is off by default and requires setting chat.autopilot.enabled to true.
Alongside Autopilot, version 1.111 ships a session-level permissions picker offering three tiers — Default Approvals, Bypass Approvals, and Autopilot — as well as agent-scoped hooks that let developers attach pre- and post-processing logic to specific agents via YAML frontmatter.
Google’s Parallel Move
The same week, Google released auto-approve mode for Gemini Code Assist’s agent, which the company internally labels “yolo mode” in its VS Code extension. According to Google’s agent mode documentation, the feature automatically permits all agent actions without requiring user confirmation for each step. Users enable it via the geminicodeassist.agentYoloMode setting in VS Code or an “Auto-approve changes” checkbox in IntelliJ.
Google’s documentation is notably blunt about the risks. It warns that “the agent has access to your machine’s file system and terminal actions as well as any tools you’ve configured for use” and urges users to “be extremely careful where and when you auto-approve changes.” A separate caution notes that changes made outside the IDE cannot be reversed, and the feature only functions in trusted workspaces. Google first introduced auto-approve mode as a preview feature in mid-2025, and continued expanding its agent capabilities through March 2026 with the addition of Gemini 3.1 Pro for agent mode.
As The Register noted, the security documentation for these features acknowledges risks including the non-deterministic nature of generative AI, prompt injection vulnerabilities, and tool poisoning attacks through the Model Context Protocol (MCP). Microsoft recommends terminal sandboxing or running VS Code in a dev container rather than relying on auto-approval rules alone, though terminal sandboxing currently only functions on macOS and Linux.
What We Don’t Know
Neither Microsoft nor Google has published data on how many developers have enabled autonomous agent modes in their editors, or whether the features have been tested in enterprise environments with production codebases. The long-term effects of weekly release cadences on extension compatibility and enterprise deployment workflows remain to be seen. It is also unclear how organizations will manage the security implications of agents that can execute terminal commands and modify files without human review, particularly in environments where code changes are subject to compliance requirements.
Analysis
The simultaneous arrival of autonomous agent modes in two major development environments signals that the industry is moving past the “copilot” model — where AI suggests and humans approve — toward a delegation model where AI acts and humans review after the fact. The tension between productivity gains and security risks is visible in the documentation itself: both Microsoft and Google promote these features while simultaneously warning users about their dangers.
The weekly release cadence for VS Code is arguably the more structurally significant change. Monthly releases imposed a natural rhythm on the editor ecosystem — extension developers, enterprise IT teams, and documentation writers all synchronized to that cycle. Weekly releases compress that timeline considerably, and the fact that AI tooling helped make the shift possible suggests a feedback loop: AI agents accelerate development, which accelerates the delivery of more AI agent features.
One developer on GitHub described the weekly release approach as “confusing and concerning,” noting that settings changes required with each weekly update create workflow burden. Whether that sentiment is an early warning or a minority view will become clearer as the weekly cadence settles in.