When AI Can Run Your Computer, Automation Means Something Different

OpenAI released a model this week called GPT-6 Astra that can use a computer the way you do — opening files, filling forms, browsing websites, writing code — but at nearly twice the speed of the previous version. This is not just a better chatbot. It is a system that can handle an entire workflow from start to finish, without a person clicking through each step.

The numbers behind it are striking. Astra scored 99.9% on ARC-AGI-3 — a test built to measure human-level reasoning — while the previous model scored just 7.8% on the same test. It also got a perfect score on a cybersecurity benchmark that mimics real-world attacks. OpenAI's president Greg Brockman said the model may mark the start of the AGI era, meaning AI that can handle general problems the way a person can.

For anyone running a business or building software, this matters because automation can now reach tasks that needed judgment, not just repetition. The kind of work that takes ten browser tabs and half an afternoon — researching, pulling data, writing a summary — is now something an AI system can handle end-to-end, with a person watching but not doing every step.

There is a real safety catch, though. Astra is the first model OpenAI has flagged as reaching a critical cybersecurity threshold — it can find and exploit security flaws in systems without being told how, step by step. That makes it powerful for defenders, but risky if misused. Getting safety right is now as urgent as the capability itself. OpenAI is rolling it out in stages, starting with vetted organisations.

If your business runs on repetitive digital work — data entry, research, document processing, customer communications — the question is no longer whether AI can handle it. The question is how to set it up and supervise it well.