Google just turned every VS Code instance into an AI server room. Jules, their coding agent, now ships as a CLI, an API, and an MCP server — meaning your terminal, your CI pipeline, and yes, your editor can all spin up cloud VMs that write code while you eat lunch. The token budget? Fifteen thousand per task on the free tier. The catch? You don't see what it's doing until it's done.
Let that sink in. Your IDE is no longer a text editor with syntax highlighting. It's a launchpad for autonomous agents that fork your repo, generate diffs, and open pull requests — all without your eyes ever touching the screen. Google calls this "always-on" development. I call it "the death of the code review." Here's the timeline: May 2025 — Jules launches in public preview, web-only. August 2025 — exits beta, gets a free tier. October 2025 — CLI drops, API goes public. January 2026 — MCP server lands with thirteen tools. Now it's everywhere. Slack, terminal, VS Code, your CI runner. One thousand five hundred daily tasks on the Ultra plan. One hundred fifty thousand to two hundred thousand tokens per task.
Spotify already merged fifteen hundred AI-generated pull requests. FIFTEEN HUNDRED. And they started with simple dependency bumps — now it's refactoring fleet management systems that span dozens of repos.
The real danger isn't that AI writes bad code. It's that AI writes code you never reviewed, in a repo you forgot you connected, with a token budget you didn't know you had. Every developer needs a personal rule: no agent touches production without a human sign-off. No exceptions. Not for speed. Not for "velocity." The moment you skip the review is the moment you ship a vulnerability with a Google-generated signature on it.