OpenAI acquired Software Applications Inc. last month—the tiny startup behind Sky, an AI assistant for Mac that can see your screen and control your apps. The team? The same people who built Apple's Shortcuts app. This isn't just an acquisition. It's OpenAI building the infrastructure to make ChatGPT do things instead of just talk about them.
Who Actually Got Acquired
Ari Weinstein and Conrad Kramer created Workflow back in 2014, which Apple bought in 2017 and turned into Shortcuts. They left Apple in 2023, founded Software Applications Inc., and spent the past year building Sky—an AI assistant that floats above your Mac desktop and actually manipulates apps on your behalf.
Sky wasn't publicly released before OpenAI bought it, but early testers described it as transformative. You could tell it to "turn this webpage into a calendar event and message my team," and Sky would parse the content, create the event, and send the message. All through natural language, without you touching anything.
The technology uses macOS accessibility APIs to understand what's on your screen and take actions across any app. It's not limited to apps with APIs or integrations. If you can do something manually on your Mac, Sky can theoretically do it for you through automation.
Why This Matters More Than It Looks
OpenAI has been trying to make ChatGPT agentic—capable of taking actions, not just answering questions. They launched ChatGPT Atlas (a web browser) days before this acquisition. They've been building integrations with third-party apps. But those are all sandboxed experiences with limited scope.
Sky gives ChatGPT native access to macOS. It can see your desktop, understand context from multiple windows, and operate apps the way you would. That's a fundamentally different level of AI integration than "ChatGPT can draft an email if you copy-paste it into Gmail."
Nick Turley, OpenAI's VP and Head of ChatGPT, said the acquisition "accelerates our vision of bringing AI directly into the tools people use every day." That's corporate speak for "we can now control your computer."
The Apple Problem
This puts OpenAI in direct competition with Apple's own AI strategy. Apple is rebuilding Siri with Apple Intelligence to do exactly what Sky does—understand on-screen content and perform actions across apps. That overhaul isn't expected until spring 2026.
Meanwhile, OpenAI just bought the people who know macOS automation better than almost anyone outside Cupertino. And they're integrating that expertise into ChatGPT, which already has hundreds of millions of users.
Apple obviously sees OpenAI as both partner and threat. They're routing some Siri queries to ChatGPT when Siri can't handle them. But they're also paying Google $1 billion to power the next Siri with Gemini. That hedging strategy looks increasingly necessary as OpenAI moves deeper into OS-level integration.
The Privacy Conversation Nobody's Having
Sky—and by extension, ChatGPT with Sky's technology—needs to see everything on your screen to work. That's the whole value proposition. It understands context by reading your windows, emails, documents, and browser tabs.
OpenAI will have to handle this carefully. Apple built its brand on privacy and runs everything through on-device processing or Private Cloud Compute. OpenAI... doesn't have that infrastructure. Sky's AI probably runs in the cloud, which means your screen content is being processed on OpenAI's servers.
They haven't detailed how they'll handle privacy yet. But any system that can "see" your screen and take actions on your behalf is inherently high-risk if compromised or misused. Even with good security, people will be uncomfortable with AI watching their desktop.
What Actually Ships
The acquisition happened in late October. OpenAI hasn't announced when Sky's technology will appear in ChatGPT. Historically, they roll out new features to Plus and Team subscribers first, then broader release later.
My guess is we'll see some version of this by early 2026. Probably starting with limited actions—"send this email," "create a calendar event," "search my files"—then expanding as they build confidence in the safety and reliability.
The hard part isn't the technology. Sky already works. The hard part is making it safe, respecting privacy, preventing abuse, and not breaking people's workflows when the AI misinterprets instructions.
The Bigger Strategic Picture
This is OpenAI's second significant acquisition in a month. They also bought Roi, an AI-powered personal investing app. The pattern is clear: they're buying products and teams that turn ChatGPT from a chatbot into a platform that does things.
Atlas for web browsing. Sky for desktop automation. Roi for financial management. Add those to existing integrations with productivity tools, and OpenAI is assembling the pieces to make ChatGPT your primary digital assistant.
That's the competition: not just building better models, but being the AI you actually use throughout your day. Google has that through Search and Android. Microsoft has it through Windows and Office. Apple has it through iOS and macOS. OpenAI needs to build it from scratch by acquiring the right pieces.
Where This Goes Wrong
There's a Clippy scenario here where ChatGPT becomes too intrusive. An AI that can see your screen and take actions sounds great until it does the wrong thing at the wrong time. Or interrupts your workflow. Or makes assumptions about what you want.
The more autonomous AI becomes, the more opportunities for mistakes. And on your actual computer—where your files, accounts, and work live—mistakes have real consequences.
OpenAI needs to nail the UX of "AI that does things for you" or this becomes more annoying than helpful. That's not a technical problem, it's a design problem. And it's harder than training better models.
My Take
I'm genuinely excited about the technology. The Sky team built something legitimately innovative, and OpenAI acquiring them means that innovation might reach millions of people instead of dying as a niche Mac app.
But I'm also wary. We're moving fast toward AI systems with significant control over our computers. The safety implications haven't been fully worked out. The privacy concerns are real. And we don't know how people will actually use this once it's available at scale.
The fact that Apple is delayed on similar features until 2026 probably accelerated this acquisition. OpenAI saw an opening to be first to market with native OS-level AI automation, and they took it. That's smart business. Whether it's smart for users depends entirely on execution.
I'll be testing this the moment it's available. But I'll also be watching carefully to see what gets sent to OpenAI's servers and how much control I actually have over what the AI can do.