OpenAI did it again. But what did they do?
Ben Hylak ● Covered by 33 sources
Opinion — commentary, not a factual news event.
OpenAI says GPT-6’s big new trick is computer use. That means agents can do more real work — and fewer people get stuck copying between tabs.
Based on reporting by Ben Hylak — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
OpenAI keeps finding the next obvious thing after everyone else has already half-seen it. First it was more compute, then ChatGPT, then tool calling, then reasoning models. Now, with GPT-6, the company’s fifth move is computer use.
On paper, that sounds small. It isn’t. A huge amount of work is still just “using a computer,” and the source makes the bigger point bluntly: that once felt uniquely human, and now it doesn’t. The people who used to live in the cracks between systems — copying, pasting, nudging, bridging gaps — are the ones most exposed.
That matters for software development first. Agents that can’t actually test things, verify things, and move through interfaces are still mostly demo material. It also matters for personal assistants, because plenty of real-world services still don’t have APIs, whether by design or by accident. If an agent has to spend longer operating the machine than a person would, the whole pitch falls apart.
The source argues that better computer use is really a broader intelligence problem in disguise. It isn’t one narrow skill; it’s a general upgrade across workflows and domains. That may also help explain why GPT-6 is described as strong at 3D modeling, alongside whatever expensive dataset went into it. OpenAI’s pattern is simple once the trick is named: take a thing humans used to do by hand, and make the model good enough to eat the gap.
My take — AI-written commentary, not fact-checked reporting
This is the part of AI that actually threatens office trivia and all the glue work nobody celebrates. The hype machine loves grand new categories, but the real power move is boring: make the model better at clicking, dragging, checking, and moving on. That’s how “assistant” stops meaning chatbot and starts meaning replacement.
Read more about this at: Ben Hylak
