Google's Gemini Mac App Is Here. ChatGPT's Had One for Two Years.

Google released a native Gemini app for Mac last week. It lets you share your screen with the AI to get help with whatever is in front of you, including local files. It is polished, it probably works, and you should pause to consider the fact that ChatGPT has had almost the exact same feature since May 2024.

Nearly two years. That is how long Google, the company that literally invented the Transformer architecture underpinning essentially every major AI system in existence, needed to ship a native desktop application.

Let us do the timeline, because I think it is instructive.

May 2024: OpenAI launches a native ChatGPT app for Mac, with screen-sharing capability built in from day one. It reads your screen, understands your context, and quietly becomes the default desktop AI assistant for a lot of people who might otherwise have defaulted to a Google product.

November 2024: Anthropic ships Claude Desktop. Different interface, same idea: a native app that can interact with what is on screen and work with local files. Another major player takes up residence in the Mac menu bar.

April 2026: Google shows up.

To be clear: Gemini is good. I am not here to tell you the new app is a failure. The AI underneath it is genuinely capable, and Google has poured enormous resources into Gemini over the past two years. But capable AI trapped in a browser tab is a fundamentally different product than capable AI in a native desktop app that can see your screen in real time. For nearly two years, two of Gemini's main competitors had that native experience. Google did not.

The question worth sitting with is: why?

Part of it is organizational. Google has more AI talent than almost anyone else on earth, and for years it managed that talent across competing internal factions: Google Brain, DeepMind, the original Bard team, the TPU hardware team, the Search team trying to integrate AI into the core product. When Sundar Pichai merged Google Brain and DeepMind in 2023, the stated goal was to reduce this friction. But merging research organizations is easier to announce than to execute. People who spent years building research culture in separate teams do not suddenly start shipping consumer products together just because there is a new org chart.

Part of it is the curse of being a web company. Google's entire business is built around browser-based experiences. Search is a webpage. Gmail is a webpage. Google Docs is a webpage. The native desktop app is a product category where Apple and Microsoft have significant home-field advantage, and one Google has never really needed to be good at. Until now.

And part of it is the execution record. The Bard-to-Gemini rebrand. The image generation controversy. The promotional video that used pre-recorded footage to suggest real-time AI performance the model could not actually deliver at launch. None of these were individually catastrophic, but they added up to a perception problem the team has been fighting against ever since. When you are playing catch-up, you cannot afford to give the narrative that kind of fuel.

None of this means the Mac app will underperform. It might actually do quite well. Google has more users embedded in the Google ecosystem: Workspace, Photos, YouTube, Android. If the Gemini app eventually integrates with all of that at the desktop level, the value proposition could legitimately outperform what ChatGPT or Claude offer natively. The three-way comparison still has Gemini trailing on some dimensions, but native desktop integration changes the math.

But there is a practical problem. For two years, the users who wanted a desktop AI assistant made a different choice. And habits are sticky. The menu bar real estate, the keyboard shortcut, the muscle memory of pressing a hotkey to ask a quick question: that behavior is now associated with a different app for a lot of people.

OpenAI understood this. It is probably why they shipped the Mac app more than a year before they had a fully polished subscription model to attach it to. They were not optimizing for revenue in that moment. They were building the habit. The reflex. The "oh, I can just ask this" instinct.

Getting people to change that instinct is hard. Ask Microsoft how long Bing has been trying.

The Gemini Mac app is real now, and it is late, and it might not matter in the long run if Google's ecosystem advantages are strong enough. The honest answer is that nobody knows yet. Desktop AI habits are still forming. Most people do not have one dominant assistant the way they have a dominant email client.

But I keep coming back to one fact. Google published Attention Is All You Need in 2017. The paper that started the Transformer revolution. The paper that OpenAI, Anthropic, Meta, and essentially every other AI lab has been building on for the past nine years.

The company that wrote that paper released a native desktop AI app in 2026.

Attention is, in fact, all you need. It just apparently took a while to pay it.