Google launches CC, a shared AI agent for families and households

Google Labs has introduced a new family-focused version of CC, turning the experimental personal productivity agent it launched in December 2025 into a shared AI agent that can coordinate information and tasks across a household.

The system supports up to six members, each of whom decides what information to share with the agent rather than automatically exposing an entire Google account to the rest of the group.

CC can combine selected emails, calendars, tasks, files and messages into a shared “Your Day Ahead” briefing, maintain household-level memory, create or update Calendar events and Tasks, and perform permissioned actions such as pre-filling registration PDFs, creating shopping lists or producing weekly meal plans.

Behind the interface, Google says each CC runs on an isolated cloud computer powered by its Antigravity agentic harness and the latest Gemini models, giving the system an execution environment rather than limiting it to conversational responses.

The launch is an early Google Labs experiment rather than a general Gemini release: access initially covers people aged 18 or older in the United States using personal Google accounts, with existing CC users receiving upgrade invitations and new users joining a waitlist.

··········

CC CHANGES THE UNIT OF AI ASSISTANCE FROM ONE USER TO A HOUSEHOLD.

The architecture combines information supplied independently by several people while preserving separate permissions for each member.

The original CC was designed around one user’s Gmail, Calendar and Drive context and generated a personalized morning briefing. Google later brought that personal briefing concept into the Gemini app as Daily Brief.

The redesigned CC addresses a different problem: household logistics are distributed across several people and several accounts, while much of the underlying information should remain private.

A school notice may arrive in one parent’s inbox, an activity schedule may be sent to another person, a travel booking may belong to a third account, and family tasks may sit in separate calendars or informal messages. The new system lets members selectively contribute relevant pieces of that information to a common agent.

CC itself receives a verified Google Account, giving the agent a distinct identity inside the group rather than making it operate implicitly through one family member’s identity.

........

System element	Current CC implementation	Practical function
Household size	Up to 6 members	Multiple people can collaborate with one agent
Agent identity	Dedicated verified Google Account	Separates CC from individual member identities
Information access	Member-selected	Each person controls what CC can see
Shared briefing	Your Day Ahead	Consolidates schedules, tasks and completed work
Execution	Isolated cloud computer	Allows multi-step agent tasks
Agent stack	Antigravity + Gemini models	Reasoning, tool use and execution
Initial availability	U.S., 18+, personal Google accounts	Early experimental rollout

........

The design can be represented as individual private accounts → selectively shared context → household CC memory → planning and execution → permissioned output or action.

The coordination layer is shared; the underlying personal data remains selectively contributed.

··········

MEMBERS CONTROL WHICH INFORMATION BECOMES PART OF THE SHARED CONTEXT.

Google has made permission management part of the normal interaction model rather than treating household data as automatically shared.

Each member can choose information that CC may use and can modify those choices later. For recurring sources, Google provides an “Auto cc” mechanism that can continuously share messages from selected senders, such as a school, sports center or travel provider.

Google also says users receive a private weekly list of newly encountered email senders that they can decide whether to share with CC. One-off information can be forwarded directly to the agent through email, Google Chat, Drive or Calendar.

CC maintains persistent memory and distinguishes between group-level information and person-specific information. A regular grocery list or preferred family restaurant can belong to household memory, while a dietary preference or timezone can remain associated with one member.

That distinction becomes technically important once an agent begins executing tasks because persistent memory can otherwise turn a piece of personal context into an assumption applied across the entire group.

The permission architecture therefore has two separate dimensions: what information enters CC and which resulting actions CC is allowed to perform.

··········

CC CAN EXECUTE HOUSEHOLD WORK RATHER THAN ONLY SUMMARIZE IT.

The isolated execution environment allows the agent to move from information aggregation into forms, documents, scheduling and other multi-step workflows.

Google lists examples including pre-filling class-registration PDFs, completing permission-slip workflows, generating school-supply lists, planning meals, checking live travel times and creating shared Docs or Sheets.

CC can also extract dates and tasks from material supplied by family members and use them to maintain shared Calendar events or Task lists. If required information is missing, Google says the agent can request the missing details and subsequently update its memory so that repeated workflows require less manual context.

For example, CC could receive a sports schedule, identify a new event, check the location, estimate travel time through the Google Maps API, add the event to a shared calendar and incorporate it into a future morning briefing.

........

Workflow	Information CC needs	Possible action	Primary control point
School activity	Shared school email or attachment	Add dates and tasks	Member chooses source
Registration	PDF + required personal details	Pre-fill form	Permission before action
Family schedule	Calendar and shared messages	Create/update events	Calendar access
Travel logistics	Locations and event times	Check live driving time	Shared itinerary/context
Shopping	Household needs or school list	Create shopping list	Shared memory

........

The architecture means CC is closer to a household operations agent than a shared chatbot. A conversational model can suggest that a parent complete a registration form; an execution-capable agent can inspect the form, retrieve permitted context, fill appropriate fields and leave the result ready for review.

That additional capability also increases the consequence of an incorrect assumption. A mistaken summary is inconvenient, while an incorrect calendar modification, form entry or shared document can propagate directly into a household workflow.

··········

THE EARLY ROLLOUT WILL TEST WHETHER SHARED AGENTS CAN OPERATE WITHOUT COLLAPSING PERSONAL BOUNDARIES.

CC introduces a multi-user version of personal AI in which useful context has to remain shareable without becoming universally visible.

Most personal assistants are structurally simpler because one person owns the account, supplies the context and receives the output. CC instead has to reason over information with different owners, different permissions and different scopes while preserving enough shared state to coordinate the household.

Google’s implementation addresses that problem through a separate agent identity, member-controlled sharing, household and individual memory, permissioned external actions and an isolated cloud execution environment.

The current limitations remain substantial. CC is still a Google Labs experiment, availability is initially restricted to U.S. adults using personal accounts, and Google has not presented the rollout as a generally available Workspace or global Gemini capability.

Its usefulness will also depend on the quality of the shared context: an agent cannot reliably coordinate a household schedule when important information remains outside the sources members choose to provide, and persistent memory can amplify stale or incorrect assumptions if they are not corrected.

The practical boundary is therefore defined by permission quality, completeness of shared context, accuracy of persistent memory and the reversibility of agent actions.

CC extends Google’s agent strategy into a genuinely multi-user environment, but the household model makes data separation and action control as important as the underlying Gemini reasoning capability.

··········

FOLLOW US FOR MORE.

·····

DATA STUDIOS

·····

[datastudios.org]