Google shipped subagents for Gemini CLI today. CLI stands for command line interface, the text-based terminal that developers use to run code and automate tasks. Until now, Gemini CLI worked as a single agent handling one task at a time. Subagents change that.

The idea is straightforward: instead of one AI doing everything sequentially, Gemini can now spin up specialized subagents and run them in parallel. You give it a complex task, it breaks it into pieces, assigns each piece to a subagent, and coordinates the results. Each subagent operates in its own isolated context with its own tools and instructions, so the main session stays fast and uncluttered while the heavy lifting happens in the background. For developers building automated workflows or working on large codebases, this is a meaningful shift in what the tool can actually handle without constant human input.

You can also build your own custom subagents, essentially creating specialist team members tailored to your project. A frontend specialist agent, a codebase investigator, a documentation writer. You define them in a simple text file, drop them into your project folder, and Gemini CLI knows to call on them automatically when the task fits. You can also call them explicitly by name using the @ symbol, the same way you would tag a colleague.

Think of it like the difference between asking one person to handle an entire project versus having a manager who can delegate to a whole team. The output is faster, and the tasks can be a lot more ambitious.

Google launched Gemini 3.1 Flash TTS today, a new text-to-speech model that gives you director-level control over how AI voices sound. TTS stands for text-to-speech, the technology that converts written text into spoken audio. It is what powers AI assistants, audiobook narration, and voice interfaces.

What makes this one different is the level of control. Using something Google calls Audio Tags, you can add simple text instructions like [enthusiastic] or [slow] directly into your script, and the model adjusts its delivery accordingly. Over 200 of these tags let you shape emotion, pace, and tone on a line-by-line basis, more like directing an actor than configuring a settings menu.

The quality benchmarks back it up. In a head-to-head quality ranking based on over 1,700 listener votes, Gemini 3.1 Flash TTS scored 1,211, landing just behind the top spot and well ahead of established players like ElevenLabs. It supports 30 voices across 70 languages, and each audio output comes with a SynthID watermark so listeners can verify it was AI-generated.

It is available now through Google AI Studio and the Gemini API at $36.60 per million characters. Some early users have flagged complaints about the rollout menus and pricing structure, but the quality itself is getting strong reviews.

Google launched a native Gemini app for macOS today. A global shortcut (Option + Space) opens a mini chat from anywhere on your screen, so you can ask a question without switching apps or opening a browser.

You can also share a specific window with Gemini to get contextual help based on exactly what you are looking at. It is a small thing that changes the daily rhythm of using AI considerably. You can download it now.