VS Code 1.111 launched Autopilot: AI agent writes, tests, and deploys on its own

VS Code 1.111 was released on March 9, 2026 — and this is not just another update with minor tweaks. Every new feature in this version is related to AI. The main innovation is the Autopilot mode, in which an agent writes, tests, and deploys code on its own, without stopping and without requesting confirmations until the entire task is completed.

What is Autopilot mode

Autopilot is not an improved autocomplete. It is a fully-fledged AI agent that works in a loop: receives a task, independently plans the steps, executes them, and corrects the result if necessary. The key feature is that the agent does not wait for the developer at each step.

It approves tool calls on its own, retries on errors, and answers intermediate questions that may arise during the work. Microsoft's lead engineer Kai Metzel stated it directly: the agent should not "idle waiting for an answer." The goal is to remove all points where the process slows down due to waiting for a human.

This is a fundamentally different model than everything that came before. GitHub Copilot offered autocomplete variants — a person chose. Copilot Chat explained code and generated functions on request — but the decisions remained with the developer.

Autopilot is the first to remove the human from the decision-making loop during task execution.

What the agent does without the developer

Autopilot covers the entire typical cycle from writing code to deployment:

• generates and edits code based on task descriptions
• runs tests and interprets their results
• fixes compilation errors and linter warnings
• calls the terminal and executes build commands
• deploys changes according to the specified pipeline

The developer sets a task — and gets a ready result. What happens between these two points, the agent decides on its own.

Is this good or bad

The author of the original material has been writing code since the early 2000s. He saw Eclipse slow down to a crawl, watched IntelliJ dominate the Java development world, and saw VS Code become the default editor for an entire generation. According to him, Autopilot is not adding a new feature. It is a change in the very nature of the relationship between a developer and a tool.

"This is — different.

This is a change in the relationship between the developer and the tool itself."

The risks are obvious. An agent that approves its own actions works fast — but if something goes wrong, the developer finds out about it afterward, not during the process. A deployed bug, an overwritten config, a prod command instead of dev — all of this becomes possible when the human is removed from the chain. Autopilot is a clear bet on speed, and every team will decide whether this bet is worth the loss of control.

What does this mean

VS Code has gone from an editor with autocomplete to an autonomous task executor — and this transition took less than two years of active AI feature development. For teams, Autopilot is a real time saving on routine cycles. But it will only work in an environment with clear restrictions and explicit rules. Without them, autopilot mode turns from an acceleration into a source of risks that are harder to catch than ordinary bugs.