Remember when ChatGPT was just an AI chatbox that sat on your desktop? That was, like, so December.
Fast forward to now–three whole months later!–and the idea of a desktop app that keeps an AI confined to a chatbox seems positively quaint. Starting with vibe-coding apps like Claude Code and Codex before going into hyperdrive with OpenClaw, AI is jumping out of the chatbox and taking charge of your PC–and by extension, glomming onto your files and personal data.
Among the big-three AI providers, Anthropic is arguably the furthest along in terms of turning the Claude app into an agentic AI command center, complete with agent-coordinating tools for coding (Claude Code) and your desktop (Claude Cowork).
Within just the past week, Anthropic shipped Claude Dispatch, which lets you remotely control Claude desktop agents from the Claude mobile app, while the new Claude Channels lets you communicate with agentic Claude Code sessions via Discord and Telegram, a feature similar to OpenClaw’s social-messaging app capabilities.
Anthropic’s aggressive Claude moves have left Google and OpenAI scrambling to catch up, and we’re now getting word that both companies are doing just that.
Google, which doesn’t even have a general-purpose desktop app for Gemini (unless you count Chrome and its Gemini sidebar), is said to be testing one now, starting on the Mac side first. While it doesn’t sound like the new Gemini desktop app will offer agentic coding or desktop capabilities (you can vibe-code with Gemini using Google’s standalone Antigravity app), it will supposedly offer a “desktop intelligence” feature that allows Gemini to see what’s on your computer screen and use that data for context, similar to Microsoft’s Copilot Vision.
OpenAI, meanwhile, is reportedly cranking out a desktop “superapp” that integrates ChatGPT with Codex, OpenAI’s agentic coding app, and the Atlas AI browser, with the new app capable of unleashing AI agents that “work autonomously on a user’s computer to carry out a variety of tasks,” such as “writing software and analyzing data.”
Now, this is all happening in the context of the meteoric rise of OpenClaw, the open-source agentic AI framework that sets loose teams of AI agents on your own computer, all working silently in the background as you give them occasional direction via WhatsApp, Telegram, or any number of other social messaging apps on your phone.
The hype around OpenClaw–which, keep in mind, was barely on anyone’s radar until just two months ago–isn’t dying down. It’s now common to hear OpenClaw described as not merely an app but rather an “operating system for AI agents,” and earlier this week, Nvidia boss Jensen Huang took it a step further, calling OpenClaw “the new computer.” Naturally, he said this while unveiling NemoClaw, Nvidia’s own “secure” take on OpenClaw.
Huang isn’t the only one proclaiming that agentic AI frameworks like OpenClaw are “the new computer.” Take Perplexity, which recently unveiled Personal Computer, an OpenClaw-style agentic AI software setup that lives on a Mac mini, or Meta’s Manus AI, which just launched the OpenClaw-esque My Computer.
All these new agentic AI systems aim to escape the bounds of mere desktop apps and intertwine themselves deeply with your files and your PC. And white lip service is still being paid to privacy and sandboxing, it’s all too easy to allow software like Personal Computer, OpenAI’s upcoming “superapp,” and even the Claude desktop app access to your most sensitive files and your PC itself.
Indeed, giving AI agents total access to your computer appears to be the point, so why not just put AI agents in charge of your PC, right?
That’s the argument, anyway, and with their latest desktop apps focusing more and more on putting AI agents on our desktops, Google, OpenAI, and especially Anthropic appear headed in that direction.
