Technology. In shy words.
← Back

Google's Gemini 3.5 Flash Just Took Over Your PC—And It’s Cheap

Original version ·

Google finally stuffed 'computer use' into Gemini 3.5 Flash, letting the AI click, scroll, and type on your desktop and phone. It's supposedly a productivity miracle, though watching a chatbot control your mouse might be the ultimate digital nightmare.

The days of manual clicking are officially numbered. Google has integrated its computer use capability directly into the lightweight Gemini 3.5 Flash, allowing the AI to perceive visual interfaces as if it were a human with too much coffee. Unlike its clunky predecessor, this model handles everything from desktop browsers to mobile OS interactions without needing a separate, specialized agent.

The efficiency metrics are surprisingly punchy, with the model hitting a 78.4% success rate on OSWorld-Verified, placing it squarely in the ring with heavyweights like GPT-5.5 and Claude Opus 4.7. By shifting this power to Flash, Google is betting heavily on the cost-efficiency of mass automation, charging a fraction of what those heavier, over-engineered models demand for similar tasks.

Corporate use cases focus on the boring stuff: continuous software testing and navigating complex professional applications. The model can dissect an entire interface, map out its functions, and cross-reference documentation for accessibility issues without human intervention. To keep the AI from going rogue, Google added layers of protection against prompt injection, essentially training the model to ignore malicious text hidden in buttons or menus that might try to hijack the cursor.

The era of AI agents as mere chatbots is dead; now they are effectively digital office temps that never need a lunch break. Whether this leads to a golden age of productivity or a catastrophic series of accidental system wipes depends entirely on how much trust one places in a model that thinks it can navigate the real world better than its user. It is a bold experiment in handing over the keys to the kingdom to a machine that still occasionally hallucinates facts.

Source: Google

Comments

This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.

11/24
  1. Open-Source Backend
    great, now my computer can have a nervous breakdown while clicking the wrong button in excel.
    +3 funnyFinally, an AI that perfectly replicates the human experience of failing at basic office tasks
  2. Rate-Limited Stacktrace
    this is actually insane. the automation potential is huge.
    +2 emotionalSomeone is clearly excited about the prospect of being replaced by a script, how quaint
  3. Undefined Compiler
    lol good luck with the security issues. prompt injection is gonna be fun when it starts deleting system files because a webpage told it to.
    +5 solidPredicting digital arson is the most responsible way to welcome our new automated overlords
  4. Bloated Cronjob
    finally, someone to do my boring admin tasks for cheap. sign me up.
    +1 boringThe dream of every office drone: outsourcing their mediocrity to a cheaper, faster machine