Gemini Spark: Tasks, Connected Apps and Business Limits

Gemini Spark is useful for recurring research and draft preparation, but Google's agent is still unavailable to work or school accounts. Now called Task in the Gemini app, it supports up to 15 tasks running at once. Start with work you can inspect before anyone depends on the result. Google's task guide.

Documentation checked in November 2026. The workflows below are proposed examples, not results from a signed-in test. The adoption criteria are account eligibility, access to the required sources, dependence on a local computer, and where a person accepts the output.

What Gemini Spark Does Today

Google Gemini Spark has become Task inside the Gemini app. Google's help page explicitly explains the rename and says existing work remains available in the sidebar. Look for the Chat and Task control beside the prompt box. Current naming and interface.

The business change is that you can assign an outcome that requires follow-through. A useful assignment says what to collect, what to produce, and what must wait for your decision. “Prepare a competitor brief from these public sources” gives you something to accept or reject. “Handle my business” does not.

The Gemini Spark Agent: Task, Schedule and Skill

An agent uses tools to carry out work. Think of this one as a concierge with an assignment, an appointment book, and written house rules. The task defines the outcome, the schedule determines when work starts, and a skill supplies reusable instructions. Google's skills documentation separates these responsibilities. Tasks, schedules and skills.

For a recurring supplier brief, the task is to prepare the brief, the schedule is the agreed weekday, and the skill contains your source standards and report format. Save those standards once the draft is useful. Repeating an unclear assignment simply produces unclear work more regularly.

How to Access Gemini Spark

Check the account before changing the workflow. Google's requirements specify a personal Google Account, age 18 or over, Keep Activity enabled, and an opt-in to Personal Intelligence, its personalization feature. Managed work and school accounts are excluded. Eligibility requirements.

Included plans: Google's help page lists Google AI Pro and Google AI Ultra. Subscription requirement.

Google lists availability wherever Gemini Apps are supported, except the European Economic Area, Nigeria, Switzerland, and the United Kingdom. The United States and Canada are on its supported-country list and are not among those exclusions. Check both the task eligibility page and the Gemini country list; general Gemini access alone does not establish Task access.

The task guide names the web app, mobile app, and Mac app. Supported Task surfaces. For a founder using an eligible personal account, a public-information research pilot is a reasonable starting point. For an employee whose work lives under a company login, buying a consumer subscription does not resolve the account restriction.

How to Use Gemini Spark

Start a bounded task, inspect its output, then decide whether to schedule it. Google's desktop instructions use the Task button below the text box. The following setup adds an acceptance check to that documented flow. Create and manage a task.

1. 1

   Write the assignment

   Open the Gemini app, select Task, and describe the desired result. Name the allowed sources, destination, and stop condition. For example: prepare a private research draft, cite the source for each claim, and stop before sharing.

2. 2

   Supply the context

   Use Upload & tools for relevant files. Include the reporting format and explain which source wins when documents disagree. Give the task only material you are permitted to use with this account.

3. 3

   Submit and supervise

   Submit the task and inspect its progress. Respond when it needs clarification or browser takeover. Treat a request for access as a decision about the underlying data, rather than a routine setup click.

4. 4

   Make useful work recurring

   After reviewing a satisfactory draft, ask for a schedule and inspect it in Scheduled. Keep the owner and acceptance criteria in the instructions so that a recurring draft does not silently become permission to publish.

Google documents time-based schedules, Gmail-filter triggers, and topic monitors. A time schedule suits a weekly brief; a Gmail filter suits a specified incoming message; a topic monitor suits a change worth investigating. These task schedules are distinct from scheduled actions in ordinary Gemini chats. Schedule types and controls.

What Keeps Running When Your Devices Are Off?

Cloud execution can continue without your computer; local browser work cannot assume that independence. Google's schedule FAQ explicitly says work runs in the cloud and can continue with the device off. Its browser instructions separately require an awake device and running Chrome for local browsing. Cloud scheduling, browser execution.

That distinction changes how you design the handoff. A research assignment using cloud-accessible sources can be suitable for overnight preparation. A job that depends on the browser session on your laptop has a local dependency. “Runs in the cloud” describes where some work happens; it does not make every needed resource available there.

Which Apps and Tools It Connects To

The connection matters less than the specific action it allows. For Task mode, Google lists Gmail, Calendar, Drive, Docs, Sheets, Slides, Keep, and Tasks. The documented actions include email drafting, calendar changes, document and spreadsheet editing, presentation creation, notes, and reminders. Shared-file edits require confirmation. Task-specific Workspace actions.

For an operator, that means a brief can draw on source material and become a working document. It does not mean the agent should own the final message, deadline, or spreadsheet formula. Decide which artifact is a draft and which artifact is the record people rely on.

Google's app directory also lists Canva, a design service; Dropbox, file storage; Granola, meeting notes; and Linear, issue tracking, as supporting tasks. Their eligibility differs: Dropbox, Granola, and Linear are currently listed as US-only and English-only. Canva has its own country and language conditions. Google's Connected Apps directory.

Other documented sources include Google Contacts, Gemini Notebook, Google Photos, Search, Finance, Flights, Hotels, Maps, and YouTube. Public search information and permissioned account content are different kinds of access. In Customize > Connectors, open the relevant app to inspect its supported actions for your account and device. Connected services and controls, task sources.

For a custom business tool, Model Context Protocol, or MCP, provides a standard way for an agent to call external tools. Google lets eligible users add a custom app using its MCP server URL through the web app. Its custom-app help currently limits this to US users in English, with use on web and mobile. Google also says custom-app write actions require manual confirmation. Custom app requirements.

A solo builder should first check whether the connector exposes the needed read or write operation. A working connection that can search records but cannot update them will not finish a workflow that depends on updating a record. The presence of an app name is insufficient evidence.

Gemini Spark Use Cases With Human Checkpoints

Delegate preparation where the result can be compared with its sources. These examples assume an eligible personal account and permission to use the material. They are deliberately scoped so that a person can detect a missing source, mistaken priority, or unsupported conclusion before it becomes a business action.

Inbox Triage for a Founder

A founder returning from travel needs a decision queue, not an agent guessing which relationships matter. Use the email-search and drafting capabilities to prepare a digest of messages that appear to require attention. Keep archiving, unsubscribing, and replying outside the initial assignment.

Suggested task:

Review messages in the Gmail label I specify. Prepare a private digest grouped into decisions needed, replies to draft, and information only. Link to the original thread for each item and explain why you placed it there. Draft suggested replies in the digest. Do not send, archive, delete, or unsubscribe. Flag unclear deadlines instead of inferring them.

Human checkpoint: The founder opens the original threads for commitments, complaints, and deadlines, then checks recipients and any promises in the proposed replies. A polished digest fails if it missed the message that changed an earlier agreement.

When to schedule it: After the categories reflect how the founder works. Ask for a regular digest with the same boundaries. If the assignment needs access to a company-managed mailbox, use a supported organizational product instead of moving the mailbox into a personal account.

Competitor Research for a Strategy Operator

A strategy operator needs to know what changed and whether it affects a decision. A useful brief separates published facts from the operator's interpretation. It should also distinguish “no change found” from “the source could not be checked.”

Suggested task:

Review the public product pages, release notes, and documentation for the competitors I name. Prepare a private brief separating confirmed changes, claims that need verification, and possible implications for our positioning. Cite the original page for every factual change. Record inaccessible sources. Do not contact vendors, create accounts, or distribute the brief.

Human checkpoint: The operator opens the cited pages for claims that could alter a roadmap, sales message, or purchasing decision. Compare the source wording with the summary. A launch announcement describing a forthcoming feature should not become a statement that customers can use it today.

When to schedule it: Use a recurring research cadence when a delayed update is acceptable. Google's monitoring guidance says topic monitors are unsuitable for time-critical work. Treat the report as research input, not an immediate market-alert system. Monitoring limits.

A Status Report for an Operations Lead

An operations lead often has the source material already. The work is reconciling status notes into an intelligible draft without turning incomplete notes into invented certainty. For a permitted personal-account pilot, use approved sample material or nonconfidential project documents.

Suggested task:

Use the source documents and spreadsheet I identify to draft the next project status report. Separate completed work, blockers, decisions needed, and unknowns. Link each status claim to its source. If a source is stale or contradicts another source, flag the conflict. Create a separate draft document. Do not change source files, sharing permissions, or send the report.

Human checkpoint: The operations lead validates blockers with their owners and reconciles any totals against the original sheet. The reviewer decides whether an omission means “nothing happened” or “the evidence is missing.” Only then should the report become the version other people use.

When to schedule it: Once the source locations, template, and report owner are stable. If a required file is unavailable, the task should report that gap rather than write a plausible replacement narrative. The payoff is a prepared document with visible uncertainty, not an unattended claim that the project is healthy.

Workspace Admin and Data Controls

Workspace controls do not remove Spark's personal-account restriction. A company can enable the Gemini app and connected services for employees while Task mode remains unavailable to those managed accounts. Keep the consumer agent and the organization's approved Gemini deployment separate in your rollout decision.

For the Gemini app used with Workspace accounts, administrators can control app availability, connected Workspace services, and activity history. The Admin console path Generative AI > Gemini app > Apps controls service connections, with settings that can apply to organizational units or configuration groups. Admin connection controls, app and activity controls.

Data protections depend on the Workspace license. Google's core-service documentation says chats and uploaded files are not human-reviewed or used to improve generative AI models. Its additional-service category has different terms and may permit those uses. The admin also controls Keep Activity and conversation-retention settings. Workspace account protections.

A personal-account pilot has a different data boundary. Google's consumer Privacy Hub says some chats can be reviewed and used to improve its services when Keep Activity is enabled. It also explains that disconnecting an app does not erase Gemini activity or data already sent to that app. Consumer data handling.

For a mid-market CTO, the practical question is whether the intended account, connected sources, and retention rules are approved together. Do not solve a missing feature by copying internal material into a personal account. If the requirement is an organization-wide agent deployment, start with the separate Gemini Enterprise guide and verify the relevant feature in your tenant.

Where It Stops, and What “Always On” Overstates

A recurring agent still needs capacity, access, and a reviewer. Google permits up to 50 active schedules and 15 concurrently running tasks. A schedule will not run when the task concurrency or compute limit is reached, and run times are approximate. Current schedule limits.

Those limits are not a throughput guarantee. A report promised for a meeting needs an owner who checks that it arrived, checks the contents, and has a fallback if the task did not finish. Do not make a draft's successful appearance the only evidence that the underlying workflow is healthy.

Gemini Spark Beta: Supervision Still Matters

Google describes task execution as experimental. Its controls distinguish permission for an individual action from permission saved across future tasks. In the app's action settings, Always ask and Always allow have different consequences; approving a broad permission can change later runs too. Task permission settings.

Use the narrowest permission that completes the pilot. A prompt saying “draft only” is useful instruction, but a business should also inspect which actions the connector is authorized to take. Google specifically notes that bulk changes to private Google Tasks can occur without confirmation. A universal “it always asks before changing anything” claim would be wrong. Google Tasks exception.

The cloud claim can also be overstated. A task may pause for login, missing context, or a human decision. An external page or message can contain instructions that try to redirect an agent, a problem called prompt injection. Review the evidence and intended destination before approving a consequential action; permission to act does not establish that the proposed action is correct. Google's task safeguards.

Gemini Spark vs Claude Cowork and ChatGPT Work

Choose by account governance and required tools before comparing the quality of a demo. Anthropic Claude Cowork is an agent workspace for carrying out knowledge work; OpenAI ChatGPT Work handles multi-step tasks using available files, apps, and tools. Both now have cloud execution, so background work alone does not decide the choice.

The comparison uses Google's task documentation, Anthropic's cloud and device guidance, and OpenAI's Work architecture and scheduled-task documentation.

Claude Cowork is worth evaluating when your output is a document or research artifact and the needed connectors are available. Avoid assuming that a cloud session can always reach a folder on a sleeping computer. See the Claude Cowork guide for the fuller workflow distinction.

ChatGPT Work is worth evaluating when the task spans apps under an approved workspace configuration. Its enterprise documentation describes cloud coordination with steps that can run locally or remotely. It also says cloud-browser website sign-in is unavailable in Enterprise and Edu workspaces, an important limit for tasks that depend on authenticated websites. Work access boundaries.

Use the ChatGPT Work review for a deeper evaluation. For either alternative, test the particular source-to-deliverable path that matters to your business. A generic ability to browse, connect, or schedule is not proof that your account can complete that path.

What to Do Differently on Monday

Adopt now if you have an eligible account and a useful, reviewable preparation task. A solo founder can start with a public competitor brief. A senior operator can design the acceptance checklist before asking the agent to repeat the work. A builder can verify the needed connector operation before investing in a custom skill.

Wait on Spark for managed company work. The personal-account exclusion is decisive for a Workspace rollout. Evaluate the supported organizational products with your administrator instead. If your work only needs an occasional answer or rewrite, an ongoing agent assignment adds little value.

On Monday, choose a recurring deliverable, identify the authoritative sources, name the reviewer, and write the stop condition into the task. Review the initial output against the originals before scheduling it. Keep responsibility for accepting the result with the person who understands its consequences.

For practical guidance on choosing and supervising business AI workflows, join the newsletter.