Google released the Gemini app for Windows on 10 September 2026, announced by Erin Pettigrew, Director of Product Management for the Gemini App. It is available globally for Windows 10 and 11, opens over whatever window you are in with Alt+Space, and is marketed as lightweight enough not to slow the machine down.
If you already run Claude, none of that is new ground. Claude has shipped native desktop apps for Windows and Mac for some time, alongside Claude Code and Cowork. So the interesting question is not which vendor has a desktop app. Both do. The question is what happens to your governance position once an assistant can sit on top of every application on a staff laptop.
What actually shipped in the Gemini Windows app
Worth being precise about the feature set before drawing conclusions from it, because the capability list is what determines the control questions you need to ask.
An Alt+Space keyboard shortcut that opens Gemini over any active window or application, without switching context.
A dedicated workspace inside the app for longer pieces of work, including handing multi-step tasks to Gemini Spark, a 24/7 personal AI agent that requires a Google AI subscription.
Drafting and summarising that pulls from Gmail and Google Drive.
Built-in image and video generation, using Nano Banana and Gemini Omni, from the desktop app itself.
General availability worldwide on Windows 10 and Windows 11 from launch day.
That is a genuine product release rather than a thin update. It is also, for a Claude shop, a description of ground Claude already occupies. Platform coverage has stopped being a differentiator between the two, which is useful, because it moves the argument onto the things that actually decide whether a rollout survives a risk review.
Is Claude or Gemini the better desktop app for Australian teams?
Platform coverage will not settle it, because as at September 2026 both vendors ship native desktop applications for Windows and Mac. For an Australian business the decision turns on three things instead: what the assistant can see across the desktop once it is installed, which of those capabilities an administrator can turn on or off centrally rather than leaving to each staff member, and whether the organisation can produce a record afterwards of what the assistant read and what it did. Answer those three and the desktop question answers itself.
Why a system-wide overlay changes the control question
A chat window in a browser tab has a natural boundary. You paste something in, you get something back, and the assistant sees what you chose to show it. An assistant that opens over the active window with a keystroke has a different shape. It is designed to be useful precisely because it does not need you to move the information to it.
That is a real productivity gain and it is also the reason procurement should slow down. The same convenience that removes copy-and-paste also removes the moment where a person decided what the tool should see. On a work laptop that might be a customer record, a payroll screen, a legal letter, or a client file that sits under an obligation your business signed years ago.
Questions to put to any vendor before a desktop assistant goes on staff machines
| Question | Why it matters |
|---|---|
| What can the assistant see when it opens over an app? | Scope of visibility decides which records are in play under the Privacy Act |
| Can IT disable specific capabilities centrally? | If control sits with each user, the policy is a suggestion rather than a control |
| Is there an admin record of what it was asked and what it read? | Without a record you cannot evidence anything to an auditor or a client |
| Which subscription tier unlocks the agent features? | Agent capability can sit behind a separate subscription, which changes both cost and approval scope |
| Where is the data processed and retained? | Residency and retention answers drive the risk assessment, not the feature list |
| How do staff turn it off in a sensitive moment? | People need a fast, obvious off switch or they will work around the tool entirely |
What we see on Australian client rollouts
Most of the effort in a desktop assistant rollout is not technical. On Sydney and Melbourne engagements the pattern is consistent: the install takes an afternoon, and the governance work around it runs four to six weeks. That covers deciding which teams get which capabilities, writing the acceptable use position in language staff will actually follow, configuring the admin controls, and setting up the reporting that lets someone answer a question about usage three months later.
For a mid-market business we generally see $15,000 to $35,000 of work in that first pass, depending on how many teams are in scope and how much existing policy can be reused. That number surprises people who have costed the licences and nothing else. It is also the part that determines whether the rollout is still running a year later or has quietly been switched off after an incident.
If you want the developer-side comparison rather than the governance one, we covered that separately in Claude Desktop versus Gemini on Mac for Australian dev teams. For the permission and approval-gate detail on the Claude side, Cowork governance, permissions and approval gates goes deeper, and audit trails for regulated Australian teams covers the evidence question.
What not to conclude from this launch
Do not read Gemini arriving on Windows as a signal to re-open a settled model decision. A platform release tells you where a vendor is investing. It does not tell you whether the assistant is better at the work your team does, how it handles your document formats, or what your people will actually adopt.
Equally, do not read Claude having been there first as a reason to skip the same scrutiny. A system-wide assistant raises identical questions whoever ships it. If your Claude desktop deployment went out without central controls and without a usage record, the fact that a competitor has now caught up on platform coverage is the smaller of your two problems.
The useful move this month is to write down the six questions above, answer them for whatever you already have installed, and only then look at what anyone else is shipping. If you want a hand doing that, see our consulting services, or read Google's original announcement on the Gemini app arriving on Windows.
