Google just announced that Gemini 3.5 Flash — their fast, efficient, widely-used AI model — now has computer use built in as a native capability. No separate tool or extra setup. The same Gemini you already use can now see a screen, understand what is on it, and take actions across browsers, mobile apps, and desktop software.
This is what the phrase AI agents actually means in practice, and for developers and businesses, it just got significantly more accessible.
Major Highlights
- Google DeepMind announced today that computer use is now a built-in native capability in Gemini 3.5 Flash, previously only available as a separate standalone model.
- This means Gemini 3.5 Flash can now take actions across browser, mobile, and desktop environments, without any additional tools.
- Key use cases unlocked: continuous software testing, automating knowledge work across professional applications, and building custom agents for long, multi-step tasks.
- Google has built in safety measures, including adversarial training to resist prompt injection attacks (where a malicious website tries to hijack the AI agent mid-task).
- Two optional enterprise safeguards are available: (1) require explicit user confirmation before sensitive or irreversible actions, and (2) automatically stop a task if an indirect prompt injection is detected.
- The capability is available now via the Gemini API and Gemini Enterprise Agent Platform, meaning developers can start building with it immediately.
- Early customers already building on it include Browserbase, Browser Use, and UiPath — enterprise automation companies that run AI across large-scale business workflows.
KINI BIG DEAL
‘Computer use’ sounds technical but the idea is actually very simple. You know how you might tell someone “go online, find the cheapest flight to Accra next Friday, and book it with my card”? An AI with computer use can do that. It can open a browser, go to flight booking sites, compare options, navigate the pages, enter your details, and complete the booking, without you clicking a single button.
That is the capability Google just made standard in Gemini 3.5 Flash. And because Flash is their fast, cost-efficient model — the one you would use for large-scale deployment, not just one-off experiments — this matters more than a feature announcement typically does.
For African businesses and developers, this is worth paying attention to. A lot of the work that fills African back offices — data entry, form-filling, cross-platform reconciliation, customer service workflows — is exactly the kind of repetitive, multi-step computer work that agents like this are built for. You do not need to build a custom robot to automate these tasks anymore. You need Gemini 3.5 Flash and a developer who knows how to use the API.
The safety angle matters too. The most obvious risk with an AI that can use your computer is that a malicious website could slip in hidden instructions and hijack what the agent is doing. Google has built countermeasures for this. They are not foolproof, and the company is transparent about that. But the fact that they are thinking about it from launch rather than patching it after something goes wrong is the right posture.
Google DeepMind — Introducing computer use in Gemini 3.5 Flash
