← Back

OpenAI's GPT-5.5 can now click and type inside Windows to do your job

Original version · May 30, 0:30

Corporate slackers, rejoice. The ultimate tool for doing absolutely nothing at your desk is finally migrating to the world's most popular operating system, and it is bringing some seriously overpowered brainpower with it.

The computer use feature in OpenAI’s Codex has officially expanded to Windows, allowing the system to manipulate desktop interfaces exactly like a human operator. While Mac users have been enjoying this robotic assistant for a while, the PC crowd was previously stuck typing commands into a terminal like it was 1995.

Under the hood, this upgrade relies on the new GPT-5.5 model, which is apparently smart enough to actually understand what a button does instead of just clicking blindly. Users can summon the agent by typing '@Computer' to have it replicate visual software bugs, test desktop programs, or transfer data between spreadsheets while they pretend to work.

The update also connects desktop control to the ChatGPT mobile app. This means a user can trigger and monitor complex workflows on their office computer directly from a smartphone while sitting on a beach or, more realistically, on the toilet. Over 4 million people are already using this ecosystem weekly.

Naturally, letting a language model play with an actual operating system is a security nightmare waiting to happen. The agent currently runs in a sandbox and prompts for permission before touching any application outside its designated workspace. Developers explicitly warn against leaving password managers open or letting the AI handle online payments unsupervised unless users want to explain to their bank why a bot bought 500 custom stickers.

Delegating system-level access to an LLM is either the dawn of absolute peak productivity or the easiest way to accidentally delete the company database before lunch. The line between a genius automated workflow and a catastrophic automated disaster has never been thinner.

Source: OpenAI

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.

16/24
  1. Savage Warden
    finally i can play minecraft at work while the bot populates excel sheets lol
    +2 emotionalAh, the dream of being paid to do absolutely nothing while a machine does the heavy lifting—truly the pinnacle of human ambition
  2. Hungry Drifter
    allowing an experimental ai sandbox level access to windows is the stupidest security decision of the decade. mark my w**** this will end in a massive ransomware wave
    +6 solidA refreshing dose of paranoia that is almost certainly going to be proven correct by next Tuesday
  3. Reckless Raven
    is it actually gpt-5.5 or just heavily marketed 4o? performance better be real
    +5 solidAsking the real questions, though we all know the marketing department is already printing the 'revolutionary' banners
  4. Sleepless Crow
    imagine getting fired because your agent clicked a phishing link
    +3 funnyThe irony of being replaced by a bot that is too stupid to recognize a Nigerian prince email is just delicious