If there’s one word that defines Google’s latest AI wave, it’s agents — models that don’t just answer, but act. Across June and into July 2026, Google shipped a stack of updates aimed squarely at letting AI see, reason, and do real work. Here’s what landed and why it matters.
Gemini 3.5 Flash Gets ‘Computer Use’
Google launched Gemini 3.5 Flash, the first in its latest series to combine frontier intelligence with action. The headline feature: computer use baked in, letting you build custom agents that can see, reason, and take action across desktop, mobile, and browser environments.
Why it matters: ‘Computer use’ is the bridge between a chatbot and a genuine digital assistant. Instead of copy-pasting between apps, an agent can navigate them for you — filling forms, pulling data, clicking through a workflow. For small businesses, that’s the difference between AI that advises and AI that executes.
Gemma 4 Brings Agents to Your Laptop
Gemma 4 12B puts smart AI agents directly on your machine, running locally in just 16GB of memory. No cloud round-trip required.
Why it matters: Local models mean privacy (your data never leaves your device), zero per-token cost, and offline capability. For anyone handling sensitive information — or just tired of usage limits — a capable local agent is a big deal.
A Managed Deep Research Agent
Google released the Gemini Deep Research Agent in preview — a managed agent that plans, executes, and synthesizes complex, multi-step research workflows on its own. Point it at a question and it does the digging.
Why it matters: Multi-step research is exactly the kind of tedious, high-value task worth automating — competitive analysis, market scans, literature reviews. This is a preview of a world where ‘go research this for me’ is a real command.
Also Worth Knowing
- Gemini Omni Flash is in API public preview — a natively multimodal model for building custom, dynamic video workflows.
- Google AI Studio now turns plain-language instructions into working Android apps.
- Gemini 3.5 Live Translate does live speech-to-speech translation across 70+ auto-detected languages.
- Heads-up for developers: the old Gemini CLI reached end-of-life on June 18, 2026, replaced by Agentic 2.0 CLI — migrate any scripts or CI/CD pipelines that used it.
The Bigger Picture
Between Google’s computer-use agents and local Gemma models — and Anthropic’s newly agentic Claude Sonnet 5 — the industry is converging on the same idea: AI that takes actions, not just AI that talks. The practical move for you is to identify one repetitive, multi-step task in your work or business and start automating it. The tools to do it just got a lot better.
Ready to build with these tools? Browse our plain-English AI guides to turn agent hype into working systems.