Google just gave gemini 3.5 flash Place your hand on the mouse.
The company says Gemini 3.5 Flash has built-in computer usage, allowing developers to build agents that can interact with browser, mobile, and desktop environments. In layman’s terms, this means that the AI can look at your screen, decide what to do next, and suggest actions like clicking, scrolling, and typing.
Gemini 3.5 Flash now works on screen
This is more than just a chatbot upgrade. computer use It is a part of AI that allows models to operate through normal software interfaces, just like humans do.
google says Gemini 3.5 Flash can support “see, reason, and act” agents across browser, mobile, and desktop environments. This is important because many companies still perform critical work within older apps, web dashboards, and management portals that don’t necessarily offer clean APIs.
Previously, Google offered computer usage through the standalone Gemini 2.5 computer usage model. Currently this functionality is inside the main gemini 3.5 flash This allows developers to easily combine screen controls with other Gemini tools.
The simple version is:
| Features | what it means | why is it important |
| Understanding the screen | Gemini reads screenshot | Agents can understand cluttered interfaces |
| UI action | Suggest click, type, scroll | Move between apps without custom APIs |
| Safety decisions | can ask for confirmation | No need to blindly perform sensitive actions |
| Rapid injection detection | Can scan for hidden hostile orders | Improve the safety of live websites and enterprise tools |
Why Google Needs Agents for Everyday Work
It’s not just the features that are interesting. That’s what Google has to say about where it thinks AI is going.
A chatbot answers your questions. The agent completes the work. This change is important because businesses need more than just smarter text generation. You need help with repetitive, multi-step tasks such as checking forms, testing software, moving information between systems, reviewing documents, and handling administrative workflows.
Google, developers and companies Gemini API and Gemini Enterprise Agent Platform. The company also introduces builders to browser-based demos, reference implementations, and developer documentation.
For South African readers, the bigger question is a practical one. Could this help banks, retailers, insurance companies, telcos, and logistics companies automate time-consuming back-office tasks without having to rebuild all their internal systems?
I think that’s what’s interesting. Many local businesses still rely on web portals, spreadsheets, and manual checks. Agents that can work safely on these surfaces can potentially save time, but only if companies control where agents click and what data they touch.
Safety is now the main story
Google knows this can cause problems.
The company says it uses targeted adversarial training to reduce the risk of immediate injections. It also provides enterprise safeguards that allow you to require explicit user confirmation for sensitive or irreversible actions and stop tasks when indirect prompt injections occur.
This is important because AI agents do more than just generate words. They may take action. Unauthorized clicks can result in form submissions, file changes, messages, or disclosure of personal data.
In Google’s own developer documentation, it’s called “Using your computer.” Preview function It also warns you that it may contain errors or security vulnerabilities. The document also advises close supervision of critical tasks and recommends avoiding the use of computers for important decisions, sensitive data, or activities where major mistakes cannot be corrected.
For South Africa, that warning rings true. Businesses that use these agents for customer records, employee files, or financial workflows should still consider POPIA. The Information Regulator said: POPIA sets out minimum requirements for the lawful processing of personal information by public and private institutions.
Google is participating in a larger competition for computer usage
Google isn’t alone here.
introduction of humanity Claude’s use of the computer public beta version In October 2024, developers will be able to tell Claude to look at the screen, move the cursor, click buttons, and type text. OpenAI then launched Operator in January 2025 as an agent that can complete web tasks using its own browser.
Google’s advantage is distribution. Gemini is already close to Search, Maps, Android, Workspaces, and Google Cloud. This gives Google a powerful entry into the habits of both consumers and businesses.
We’ve already seen this direction Breakdown of Gemini Intelligence on Android by MemeburnHere, Google has embedded AI even deeper into everyday device actions. This new Gemini 3.5 Flash update brings the same idea closer to developers and enterprise automation.
What are you looking at next?
The next test is reliability.
Can Gemini 3.5 Flash complete long workflows without drifting? Can it handle messy South African websites, bank portals, government forms, internal tools, etc.? Can it pause at the right time before doing something dangerous?
While this functionality seems powerful, the best use cases may start small. Think software testing, form checking, data entry, document review, and workflow monitoring. These are areas where humans can stay in the loop while the AI takes care of the boring clicks.
We think the real story here is not that AI can use computers. That means Google wants its AI agent to become regular business software.
And where should companies draw the line when agents can click and use the same tools you use every day?
FAQ
Gemini 3.5 What is computer usage with Flash?
Let’s use the computer gemini 3.5 flash Interact with the screen through actions such as clicking, typing, and scrolling. Developers must build an execution environment to perform these actions.
Can regular Gemini users access this?
This is mainly developers and companies Through the Gemini API and Gemini Enterprise Agent platform. This is still not a regular Gemini app feature that everyone can use.
Is a Gemini computer safe to use for sensitive work?
Google includes a safety tool, but its documentation still calls it “Using Your Computer.” Preview function. When it comes to sensitive data, payments, and legal actions, businesses need to manage humans.
