Short answer. Google announced on September 23, 2026 automation of actions between apps on enterprise Android devices and control over it through policies. But its current admin help draws an important boundary: agentic services are not allowed to open and automate apps in the work profile. The automation controls described apply to fully managed devices and the personal profile on corporate devices with a mixed-use setup; in the help, this is marked as an Android 17+ feature. So the promise that “an agent will carry out actions in work profile apps on its own” cannot be considered confirmed by this announcement.
This material is for IT leaders, EMM administrators, and teams planning mobile AI use cases. Below is what the sources actually say, how the management modes differ, and what to check before a pilot. The availability of any specific feature, device model, EMM, subscription, and region must be confirmed in your own environment.
What Google announced on September 23
In the Android Enterprise blog Google described Gemini using screen, calendar, email, and browser context for multi-step actions across apps. It also named three guardrails: separation of work and personal profiles, the ability for the IT team to configure or disable automation across the device fleet, and preservation of existing device management rules. The announcement combines already available tools and plans for 2026; it does not include an availability matrix by country, EMM, or device model.
The wording “announced” matters. The post shows the product direction, but by itself it does not confirm that the needed scenario already works on every company phone. In the same article, Google separately marks some other capabilities as expected later in 2026. A rollout decision requires current documentation and device testing, not just a news post.
Work profile, personal profile, and fully managed phone
Work profile is a separate space for company apps and data on an Android device. On an employee’s personal phone, the organization manages the work profile but cannot see personal data; on a corporate device with mixed use, it can also apply some rules to the whole device. A fully managed device is dedicated to work and is managed by the organization end to end. This separation is described in Android Enterprise documentation.
| Mode | Where the apps are located | What Google help says about agentic automation |
|---|---|---|
| Personal phone with a work profile (BYOD) | Work apps are isolated from personal apps | Agentic services cannot automate work profile apps |
| Corporate phone with work and personal profiles (COPE) | Both profiles exist on one device | The automation policy may apply to the personal profile; the work profile is closed to agentic services |
| Fully managed corporate device | Corporate apps run in a managed whole-device environment | The help indicates that agentic automation can be controlled by policy; actual availability must be verified |
This is a restatement of Google's admin controls table, not a promise that every EMM already shows a new toggle. The help names the NearbyAppStreamingPolicy policy for fully managed devices and the COPE personal profile; it also states directly that agentic automation must not access work profile apps. A footnote places these features and controls in Android 17+.
Gemini in the work profile: web access is not the same as app automation
There is another distinction that is easy to lose in the discussion. Android Enterprise help recommends opening Gemini in a supported browser and signing in with a work account for the work profile. It also clarifies that the Gemini mobile app is not available inside the work profile. On a fully managed device, by contrast, the help suggests installing the Gemini app and signing in with a work account if the administrator has granted access and subscription requirements are met.
A user can read a Gemini response in the work-profile browser, but that does not mean an agent is allowed to control adjacent work apps. These are different capabilities and different access boundaries. A team that needs a helper for search and drafting may find web access sufficient. A team that wants automatic updates to a case in a corporate app needs a separate check of the allowed action path, including app permissions, human confirmation, and result logging.
What the IT team should check
Google lists several classes of administrative controls: Gemini access through Google Admin, the list of allowed apps in Managed Google Play, the assistant’s access to screen content, screenshot blocking, browser restrictions, and the agentic automation policy for eligible device modes. The help separately warns that this is not an exhaustive list and that some methods depend on Android version and manufacturer.
Do not turn one toggle into a blanket conclusion that “AI is off” or “data is protected.” The Android Management API specification shows that data exchange between profiles, text copying, and contact access are governed by different rules. Even when direct agent action in the work profile is blocked, the project still needs to define separately which information an employee may manually pass to a personal profile or an external service and which company rules allow that. An agent management policy does not replace checks for user permissions, apps, browsers, or data sources.
Minimum pilot checklist
- Document the device and mode. Record the model, Android version, BYOD/COPE/full management, EMM, and the account in use. Without this, you cannot compare tests across different phones.
- Name the action. For example: “find an email, draft a reply, suggest a case change.” Separate reading, preparation, and actual writes to the system.
- Specify the profile for each app. Note where email, CRM, browser, and Gemini are opened. If the target app is inside the work profile, do not assume automation of it in the expected outcome.
- Test the settings on a test device. Compare permissions in Google Admin and EMM, then actually repeat one allowed and one blocked scenario. Save the date, versions, settings, and observed results; the current article does not replace that test.
- Limit the impact. Until behavior is confirmed, use test data, drafts, and manual approval for actions that change customer records or send messages.
This is an editorial validation plan derived from product documentation. It is not a security certification and does not promise compatibility for any specific configuration.
What we still do not know
Official materials do not provide a verified availability table for the agentic feature across Russian organizations, device models, EMM providers, and pricing tiers. Those sources also do not publish independent measurements of accuracy, execution time, error rates, or business impact. Google’s materials describe the product capability and its limitations, not the results of a pilot at a Russian company.
There is also a timing risk: the help center may be updated faster than a news release. As of the review date, September 24, 2026, the English-language help documentation includes a section on agentic scenarios and the work profile limitation. The public Russian-language help page did not show that section at the time of review. Before making a purchasing or rollout decision, recheck the current English documentation and your device fleet settings.
Frequently Asked Questions
Can Gemini agent tap buttons in work profile apps?
According to the current Google help documentation, agent services are not allowed to access or automate work profile apps. You should not promise that scenario based on the general announcement alone.
Can you use Gemini with a work account inside a work profile?
Yes. The help documentation describes access through a supported browser and the Gemini web experience with an eligible work account and administrator approval. It marks the Gemini mobile app inside a work profile as unavailable.
Is Android 17 enough to run the agent?
No. Android 17+ is listed as a prerequisite for the described agentic features and controls. You also need the appropriate management mode, policies, Gemini access, and confirmation of actual device behavior; geography and EMM compatibility also need to be checked.
Can an administrator turn off agentic automation?
Google describes control through policy for fully managed devices and the personal profile in COPE. You need to verify the exact setting location and whether it is available in your EMM with the vendor and on a test device.
Does profile separation mean an employee cannot send work data outside the company?
No. Rules for data sharing, clipboard use, the browser, and apps are separate. The organization needs to check the full data transfer chain, not just agent service access.
How can AIrassvet help with a pilot like this?
AIrassvet works with process audits, enterprise AI systems, and interface automation. For a mobile scenario, it makes sense to start with three specific steps: map apps and data by profile; define one verifiable workflow and permissions at each step; and prepare a test with acceptance criteria and manual confirmation of changes. This is a research and integration plan, not a statement of ready-made support for Android Enterprise or a specific EMM. The first step is to choose one process, its current sequence, data sources, and acceptance criteria. Discuss the project.
Conclusion
Google’s September announcement broadens the conversation about AI agents on corporate phones, but the access boundary has to be read in the documentation. A work profile remains a separate environment: using the Gemini web version with a work account and telling an agent to automate a work app are different scenarios. A pilot should be built around a specific device, Android version, app profile, administrative policy, and observable result.