Google has shipped a significant update to Gemini Spark. There is now a native macOS app, connections to a longer list of third-party apps, and real-time topic tracking that keeps an eye on subjects you care about. For an Australian business owner weighing up an AI assistant that lives on the desktop rather than inside a browser tab, this is worth paying attention to, because it puts Gemini Spark and Claude Cowork in the same category for the first time. Both are now connected desktop assistants that can act on your files and your tools. Here is a clear-eyed look at how they compare, and where each one earns its keep.
What Gemini Spark can do now
The headline change is that Spark runs on the Mac desktop and can act on your local files, not just chat about them. Google's own example is turning hours of manual file sorting into a single request: ask Spark to sort every PDF in your Downloads folder into the right place, and it does the filing for you. It also bridges the desktop and Google Workspace, so you can ask it to build a budget spreadsheet from the latest invoices sitting on your computer, then set a schedule so the sheet updates itself on a regular basis. That combination of local files, cloud tools and a recurring schedule is what moves it from a chatbot to an assistant.
On connections, Google has widened the supported list to include Google Tasks, Google Keep, Canva, Dropbox, Instacart, OpenTable and Zillow Rentals, and it is rolling out support for custom Model Context Protocol servers so you can wire in apps of your own. There is one caveat that matters a great deal for readers in Australia: the macOS version is a beta limited to Google AI Ultra subscribers aged eighteen and over, and it is starting in the United States. So for most Australian businesses today, getting access is the first hurdle, well before the feature list becomes the deciding factor.
Where Claude Cowork sits
Claude Cowork is the desktop side of the Claude app, built for people who are not developers. It reads and writes files in a folder you choose, connects to your tools through the same Model Context Protocol standard that Google has adopted, runs reusable skills, and can carry out scheduled tasks on its own without you sitting there. That is the same shape of product as Spark: a connected desktop assistant that takes actions and runs recurring work, rather than a chat window that only answers questions and leaves the doing to you.
The practical difference is less about a feature checklist and more about what you intend to build with it. If you want an assistant that files documents, drafts replies and pulls together a quick summary, both tools will handle that comfortably. If you want to hand off a repeatable process, say reconciling supplier invoices every Monday morning and flagging the odd ones for a person to review, the depth of Claude's skills and scheduled-task model is where we have seen the most durable value for small teams. It is the difference between saving a few minutes and removing a weekly job from someone's plate.
The axes worth comparing on
An Australian business owner does not need a spec sheet. A handful of honest questions will separate the two faster than any feature grid:
Which apps connect, and how deeply the assistant can read and act inside them, not just link out to them.
Whether it only answers questions, or actually takes actions and runs automations while you get on with other work.
Data handling and governance: where your information travels, how long it is kept, and who is able to see it.
Whether it does scheduled or proactive work on its own, or only responds when you prompt it.
Availability and price in your market. A United States beta on a premium subscription is a very different proposition from something a Sydney or Melbourne business can put to work this week.
Data and governance, the Australian angle
For any business handling customer records, the real question is not what the assistant can do, but where the data travels once you connect your accounting file or your CRM. Under the Privacy Act, you remain accountable for how personal information is handled by the software you adopt, and that includes an AI assistant reading files on a staff member's laptop. Both Google and Anthropic publish enterprise data terms covering retention, whether your content is used for training, and the admin controls available to you. Read those terms before you connect anything sensitive, and check that data residency and retention actually match your obligations. This is the same advice we give a client whether they land on Claude or on a Google tool, because the accountability sits with the business, not the vendor.
An honest call
If your business already runs on Google Workspace and you can get into the beta, Spark is the path of least resistance, and its fit inside the Google ecosystem is a real strength worth weighing. If your advantage comes from building automations and agents that take actions across several tools, Claude's connector, skills and scheduled-task model is the stronger base to build on, and it is the one we scope and deliver in Australian dollars. A focused Claude Cowork setup for a small team starts from about $3,500, which buys a working automation you own rather than a subscription and a blank screen.
Not sure which one fits the way your team actually works? Book a short call and we will map it against your tools and your obligations before you commit to either.
