Gemini Task mode is the clearest sign yet that Google wants its desktop app to do work on your computer, not just talk about it. A Gemini desktop update spotted by TestingCatalog on 24 September 2026 adds a switch between Chat and Tasks. Trusted testers are starting to receive the Tasks side, which is where Google’s computer use feature lives.
Two details in the build matter more than the switch itself. Tasks carries the same Gemini Live entry point as Chat, so a user could one day direct a computer use job by voice while it runs. And the computer use settings now include a menu of allowed and disallowed apps, which gives users a first real say over what the agent may touch.
None of this is live for the public yet. TestingCatalog says it does not have access to the mode itself, and a production rollout could take “weeks or longer, depending on testing.” This article sets out what the build shows and how it connects to Gemini 3.8 Live and the new Windows app. It ends with what a business should decide before Gemini Task mode reaches its staff.
Table of contents
- What TestingCatalog Found in the Gemini Task Mode Build
- From a “Task” Option to Gemini Task Mode: Five Weeks of Traces
- Why Gemini Live Inside Gemini Task Mode Matters
- Allowed and Disallowed Apps: The New Computer Use Settings
- Projects From a Folder and the Local-File Story
- Gemini Task Mode on Windows and Mac
- How Gemini Task Mode Compares With Claude and ChatGPT Agents
- What Businesses Should Do Before Gemini Task Mode Arrives
- The Gemini 4 Rumours Running Alongside
- Gemini Task Mode FAQ
- References
What TestingCatalog Found in the Gemini Task Mode Build
The report is short, so it is worth reading closely. TestingCatalog’s Alexey Shabanov wrote that Google “has shipped a new Gemini desktop app update that moves its Computer Use functionality into a new testing phase.” That phrase, a new testing phase, is the headline. Computer use has been in the code for weeks. What changed is that a small group outside Google can now switch it on.
A Chat and Tasks switch
The visible change is a switch between Chat and Tasks. Chat is the generative AI assistant people already know: you type or speak, and Gemini answers. Tasks is the mode that hands Gemini your mouse and keyboard. Putting the two side by side as equal modes tells us how Google sees the product. Gemini Task mode is presented as a second way of using the app, not as a buried setting.
Trusted testers first
Google calls its early-access group trusted testers, and TestingCatalog says those testers are “beginning to receive access to Tasks mode.” That is Google’s usual path for risky features: a small, vetted group, then a slow widening. For an agent that can open apps and change files, a cautious start is sensible, and it means the feature’s shape can still change before a wider release.
What TestingCatalog could not see
The report is honest about its limits. “TestingCatalog does not yet have access to the mode itself,” it says. So the findings describe the switch, the settings and the entry points, not how well Gemini Task mode actually performs. Nobody outside the test group has published a hands-on account yet. Treat everything below as a description of controls and plumbing, not a verdict on quality.
From a "Task" Option to Gemini Task Mode: Five Weeks of Traces
This is not the first time computer use has surfaced in the Gemini desktop app. On 18 August 2026, TestingCatalog reported the first traces. It said Google had “finally started working on computer-use functionality,” which would let Gemini “control other apps, access files in selected folders, and perform tasks similar to those currently supported by Claude and ChatGPT.”
The 18 August report
The August build described computer use as a “Task” option in the prompt-bar selector. It would sit alongside the existing controls for generating images and videos, rather than as a toggle in the navigation bar. It also listed three practical features. There was a mini-view of the active screen so users can follow along, a control to stop the device sleeping mid-task, and an optional automatic backup of selected folders to Google Drive before a task starts.
What moved by 24 September
Five weeks later, the single “Task” option has grown into a full mode with its own switch. Voice is on the table through the Gemini Live entry point, and permissions are finer-grained through the app allowlist. The table below lines up the two reports so the change is easy to see.
| Feature | 18 August report | 24 September report |
|---|---|---|
| Entry point | A “Task” option in the prompt-bar selector | A switch between Chat and Tasks |
| Who can use it | Traces in settings; possible limited testing | Trusted testers beginning to receive access |
| Voice | Not mentioned | Gemini Live entry point inside Tasks |
| App permissions | Access to files in selected folders | Menu of allowed and disallowed apps |
| Watching the agent | Mini-view of the active screen | Not re-reported |
| Safety net | Optional Google Drive backup of selected folders | Not re-reported |
| Projects | Not mentioned | Start a project from a folder, now more prominent |
“Not re-reported” does not mean removed. The September piece focuses on what is new, so the mini-view, the sleep control and the Drive backup may well still be there. They simply were not described again.
The milestones in days
The pace is quick. Measured from the first computer use traces on 18 August, the Windows app launched 23 days later and Gemini 3.8 Live 28 days later. Gemini Task mode reached trusted testers 37 days after the traces first appeared.
Why Gemini Live Inside Gemini Task Mode Matters
The Gemini Live entry point is the detail that makes this more than a catch-up feature. TestingCatalog notes that Tasks, “like regular Chat, includes an entry point to Gemini Live.” Its reading is that users “could eventually operate Computer Use tasks through real-time voice, asking Gemini to work with their laptop while maintaining a live conversation.”
Talking to an agent that is working
Most computer use agents today follow a type-and-wait pattern. You describe a job, the agent works, and you check the result. A voice channel changes that. You could correct Gemini mid-task (“not that folder, the one called Q3”), ask what it is doing, or add a step without starting over. For long jobs, talking to the agent while it works may matter more than raw speed.
What Gemini 3.8 Live adds
The timing fits. Google released Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking on 15 September. Google says the models “execute tools and API calls in the background while continuing the conversation.” That is exactly what a voice-driven Gemini Task mode would need. TestingCatalog made the same link, calling the model’s asynchronous function calling “a natural fit for this experience.”
Background tool calls and progress narration
Google’s announcement describes Extended Thinking using “early verbal cues like ‘Let me check that…'” and “live progress narration to walk users through multi-step background tasks as they progress.” In a chat window, that is a nice touch. On a computer use agent, narration becomes a safety feature, because it tells you what the agent is about to do while you can still stop it. Google’s own benchmark figures for the Extended Thinking model are below. They come from Google and have not been independently checked.
The gap between the top and bottom bars is the useful part. A model can understand speech almost perfectly and still finish only about a third of the tasks on a harder benchmark. For a voice-run Gemini Task mode, the task-completion numbers are the ones to watch, not the audio-quality scores. Our earlier look at the Gemini 3.8 Live slugs on Google’s cloud quota page tracked the model before launch.
Where the Live models are rolling out
Google’s launch post splits the rollout by audience. The consumer line is the relevant one here: Extended Thinking is rolling out in Gemini Live for everyone, which is the Live experience a desktop user would reach from Gemini Task mode.
| Audience | Gemini 3.8 Live | Gemini 3.8 Live Extended Thinking |
|---|---|---|
| Developers | Gemini API and Google AI Studio | Gemini API and Google AI Studio |
| Enterprises | Private preview in Gemini Enterprise | Private preview in Gemini Enterprise; Workspace business customers coming soon |
| Everyone | Search Live | Gemini Live; Docs for AI Pro and Ultra; Gmail and Keep for all Google AI subscribers |
Allowed and Disallowed Apps: The New Computer Use Settings
The second new element is permissions. TestingCatalog reports that “a new menu lets users configure allowed and disallowed apps, providing control over which applications Gemini can operate.” For anyone thinking about deploying Gemini Task mode at work, this is the most important sentence in the report.
What an allowlist can and cannot stop
An app allowlist answers one question: which programs may the agent drive? That is a real control. It can keep Gemini out of a banking app, a password manager or an HR system. It does not answer a second question: what may the agent do inside an app it is allowed to use? If a browser is on the list, everything reachable through that browser is in scope, including webmail, cloud storage and admin consoles.
Why a denylist alone is weak
The menu reportedly offers both allowed and disallowed lists. The difference matters. A disallowed list blocks what you remember to name. An allowed list blocks everything you did not approve. For a business, the safer default for Gemini Task mode is a short allowlist, with new apps added on request. It is the same principle behind least-privilege access in data protection work.
Drive backups and the large-folder problem
The August build’s optional backup to Google Drive is a sensible safety net, and TestingCatalog noted it “could provide a useful safety net for people who do not use Git or another version-control system.” It also flagged the catch: “If a selected folder contains hundreds of gigabytes, uploading it to Google Drive could take a long time and consume substantial storage before the task can even begin.” For a company, there is a third question. A backup is a copy of local data in the cloud, so it needs to follow the firm’s data-handling rules.
| Control | What it does | Risk it reduces | Open question |
|---|---|---|---|
| Allowed apps | Limits Gemini to named programs | Agent wandering into sensitive apps | Can admins set it centrally? |
| Disallowed apps | Blocks named programs | Known high-risk apps | Does it cover web apps in a browser? |
| Mini-view | Shows the active screen while Gemini works | Silent mistakes | Is there a one-click stop? |
| Drive backup | Copies selected folders before a task | Unwanted file changes | Storage use and data-residency rules |
| Sleep prevention | Keeps the device awake during a task | Half-finished jobs | Battery and screen-lock policy |
| Live voice | Lets users steer a running task | Late corrections | Does a spoken “stop” halt it at once? |
Projects From a Folder and the Local-File Story
A quieter line in the report may matter as much as the voice link. TestingCatalog says “the previously spotted option to start a Gemini project directly from a folder has also become more prominent and appears to be entering trusted testing.”
Folders as the unit of permission
Starting a project from a folder turns a directory into a workspace. Gemini gets the files in that folder as context, and the folder becomes a natural boundary for what a task may change. Pair that with the Drive backup of selected folders, and a pattern emerges. Google seems to be building Gemini Task mode around folders as the unit of trust: pick a folder, back it up, let the agent work inside it.
How it links to Gemini Spark
Folder access is not new to Gemini’s desktop plans. Google’s download page for Windows lists connecting “local folders with Gemini Spark to organize and synthesize your files” as a Mac feature. We covered the folder angle in detail when the Gemini desktop Obsidian integration surfaced. Gemini Task mode looks like the step where folder access stops being read-mostly and becomes read-and-act.
Gemini Task Mode on Windows and Mac
Gemini Task mode is arriving on a desktop app that is itself very new on one of its two platforms. TestingCatalog notes that “the Windows app only launched publicly earlier this month.”
The Windows app is two weeks old
Google launched the Gemini app for Windows on 10 September 2026 for Windows 10 and 11. Pressing Alt + Space opens Gemini over the active window. Inside the app, users can hand off multi-step tasks to Gemini Spark, which Google calls “your 24/7 personal AI agent,” and create images with Nano Banana. Google said it is “just the beginning for the Gemini app on Windows, with more native desktop capabilities rolling out over time.” Our review of the Gemini app for Windows covered the launch and its gaps.
Workspace admin controls
For companies, the Workspace update is the part to read. Google says the Windows app is “ON by default for all organizations with Gemini enabled” and is “controlled by the Generative AI settings in the Workspace Admin console.” There is “no end user setting for this feature.” Whether Gemini Task mode will get its own admin switch, separate from the app as a whole, is one of the most important unknowns. Admins will want to allow chat while holding back computer use.
Features the Mac got first
The Mac app arrived earlier, in April 2026, and Google’s download page lists Mac-only features: sharing screen context by pressing both Command keys, dictating “clean text” across open apps, and the Spark folder link. Google has not said which platform will get Gemini Task mode first. The TestingCatalog report does not name one either, so it is unclear whether trusted testers are on Mac, Windows or both.
How Gemini Task Mode Compares With Claude and ChatGPT Agents
TestingCatalog’s August report framed computer use as Google closing a gap, describing tasks “similar to those currently supported by Claude and ChatGPT.” Google is not first here. Anthropic publishes a computer use tool for developers, and both rivals’ desktop apps already let agents act on local apps and files in some form. Our report on Claude’s Cowork browser in the desktop app shows how far that has gone.
Same idea, different starting points
What could set Gemini Task mode apart is the combination, not any single feature. A computer use mode paired with a real-time voice model built for background tool calls would be an unusual pairing. Google also has distribution: Workspace admins can switch the whole app on for an organisation. The table below compares the two modes inside Gemini’s own app, which is the choice users will actually face.
| Question | Chat | Tasks |
|---|---|---|
| What Gemini produces | Answers, drafts, images and videos | Actions in apps and files on the device |
| Who does the clicking | You | Gemini |
| Gemini Live entry point | Yes | Yes, per the September build |
| Main risk | Wrong information | Wrong actions |
| Key control | Checking the answer | Allowed apps, backups and live oversight |
| Availability | Everyone with the app | Trusted testers only |
What Businesses Should Do Before Gemini Task Mode Arrives
A rollout “could still take weeks or longer,” so there is time to prepare. Four decisions are worth making now, before Gemini Task mode lands on staff laptops through a routine app update.
Decide which apps belong on the allowlist
Start with a short list of low-risk apps where an agent could save real time: a spreadsheet, a notes app, a file manager. Exclude anything that moves money, holds personal data or grants admin rights. Write the list down now, so that when the allowed-apps menu arrives you are applying a policy rather than inventing one under pressure. Teams already running autonomous AI agents will recognise the approach.
Treat Drive backups as a data transfer
If the Drive backup ships as described, every backed-up folder becomes a copy of local data in Google’s cloud. Check whether that is allowed for the folders your staff would choose. Client files, regulated records and anything under a data-residency clause need a clear answer before the first backup runs. Also check storage: several large project folders could fill a shared Drive quota quickly.
Write a voice-task policy
Voice control adds a new kind of instruction. A spoken request in a busy office is easier to mishear, and easier for someone nearby to give. Decide whether voice-started tasks are allowed at all at first. If they are, limit them to the same allowlisted apps and require a typed confirmation before anything is deleted or sent.
Pilot on a spare machine
The safest first test of any computer use agent is a machine with nothing on it you would miss. Give Gemini Task mode a copy of a real folder, a realistic job and a time limit, then compare the result with what a person would have done. Log where it hesitated, where it guessed and where it asked. That record will tell you more than any benchmark, and it gives the workflow automation team a baseline to judge later versions against.
The Gemini 4 Rumours Running Alongside
The same TestingCatalog report closes with a separate thread. “Some testers have reported responses apparently associated with a Gemini 4 Pro checkpoint while using Gemini 3.8 Flash, although the exact model identity remains unclear.”
What testers report
TestingCatalog adds that it “has previously spotted references to the Gemini 4 family.” It remains unclear which checkpoint Google plans to release under that name. We tracked the earliest signals in our piece on Gemini avatars and the first signs of Gemini 4 in August.
Why it’s unconfirmed
Models misreport their own names often, and a response “associated with” a checkpoint is not proof of one. Google has announced nothing. The practical point for Gemini Task mode is simple: the agent’s quality will depend on the model behind it. If a Gemini 4 model arrives before a wide rollout, the computer use experience testers see now may not be the one the public gets.
Gemini Task Mode FAQ
Is Gemini Task mode available now?
Not to the public. According to TestingCatalog, trusted testers are “beginning to receive access” to Tasks mode in the Gemini desktop app. Google has made no announcement, and a wider rollout could take weeks or longer.
Will Gemini Task mode work by voice?
It may. The Tasks mode includes the same Gemini Live entry point as Chat, which suggests voice control of computer use tasks. Google released Gemini 3.8 Live, a voice model that runs tools in the background during a conversation, on 15 September.
Can I stop Gemini using certain apps?
The new build includes a settings menu for allowed and disallowed apps, which controls which applications Gemini can operate. Whether Workspace admins can set that list centrally has not been reported.
Does Gemini Task mode work on Windows?
The Gemini app has been available for Windows 10 and 11 since 10 September 2026, and on Mac since April. Neither Google nor TestingCatalog has said which platform the Tasks test is running on.
Will Gemini Task mode cost extra?
Google has not said. Some Gemini agent features, such as the Spark beta on Mac, launched first for Google AI Ultra subscribers, so a paid-tier start would not be surprising. That is a guess, not a report.
References
Gemini Task mode in testing along with Gemini Live support (TestingCatalog)
Google tests Computer Use on Gemini desktop (TestingCatalog)
Introducing Gemini 3.8 Live and 3.8 Live Extended Thinking (Google)
The Gemini app is now available for Windows (Google)
The Gemini desktop app is now available for Windows (Google Workspace Updates)
Computer use (Gemini Enterprise Agent Platform documentation)
Go Live with Gemini in Chrome (Gemini Apps Help)
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.
