If you've been following the AI agent race, the acquisition of Sky by OpenAI is the most significant move on the chessboard. Sky was an AI assistant for Mac built by the team that created Apple Shortcuts, and its technology gives ChatGPT the ability to literally see your screen and control your applications.
This fundamentally shifts ChatGPT from a chatbot that talks about things to an autonomous agent that actually does them.
From Text Generation to Task Execution
OpenAI’s goal is to make ChatGPT agentic—capable of taking actions. Previous efforts were sandboxed with limited integrations. Sky breaks the sandbox by using macOS accessibility APIs to operate across any app on your desktop.
Imagine telling ChatGPT: "Take the data from this Excel sheet, analyze the trends, and draft a summary email to my team in Gmail." Sky’s technology allows the AI to parse the Excel window, understand the data, and then manipulate Gmail, all through natural language.
My friend who tested an early agent tool said, "It's like having a hyper-competent intern who lives inside your computer and can see every window you have open." That capability is massive.
The Head-to-Head with Apple Intelligence
This acquisition puts OpenAI in direct, fast-moving competition with Apple's own AI strategy. Apple is rebuilding Siri with Apple Intelligence to achieve similar OS-level automation, but that overhaul isn't expected until spring 2026.
OpenAI bought the team with unparalleled expertise in macOS automation and is integrating that into ChatGPT immediately. This is smart business: exploiting a competitive lag to be first to market with native OS-level AI on a major platform.
Control and the High-Risk UX
The value proposition of this technology is that it sees everything on your screen to understand context—emails, documents, and browser tabs. This is also the central privacy nightmare. Sky's AI likely runs in the cloud, meaning your screen content is being processed on OpenAI's servers.
The hard part isn't making the technology work—Sky already does. The hard part is building the safety, privacy, and UX layers to prevent abuse and mistakes. An AI that can control your computer means that a mistake has real consequences for your files, accounts, and workflows.
My Take
I'm excited about the genuine innovation this acquisition brings. The ability to automate multi-app workflows through natural language is the ultimate productivity leap.
But I’m deeply wary of the unchecked power. We are moving incredibly fast toward AI systems having significant control over our computers, and the safety and privacy implications have not been fully worked out. OpenAI needs to nail the UX of control—making sure the user always knows what the AI sees and what it's about to do—or this will feel more intrusive than helpful.