Google I/O 2026 introduced Gemini Spark, an always-on personal AI agent that reads your Gmail and Drive and takes actions across the Google surface. It's the most ambitious consumer AI agent ever shipped by a hyperscaler, and the demos were genuinely impressive — Spark summarizing inboxes, finding contracts buried in old Drive folders, chaining actions across Calendar and Mail.

For day-to-day productivity inside one person's Google account, Spark is a real upgrade. For the documents your family actually shares — wills, insurance policies, tax slips, medical records — three structural gaps are worth knowing about before you treat Spark as the answer.

1. Spark is single-account by design

Spark is bound to an individual Google account. There's no concept of a shared family workspace where Mom uploads her will once and her three adult children can each ask questions about it from their own logins. Google Workspace has shared workspaces, but those are designed for businesses, with admin overhead and pricing that reflect that.

That mismatch matters because families don't actually use documents the way Spark assumes. When something happens — a parent goes into hospital, a will needs to be located, an insurance policy needs to be read — it's rarely the person who originally uploaded the file who needs the answer. A family document vault that fragments into one-account-per-person is a vault that doesn't work when it counts.

2. No PII redaction between your documents and Google's models

When Spark answers a question about your tax slip, the language model sees the raw document. Names, SIN numbers, account numbers, addresses — all of it goes into the model's context window. Google's policies cover what happens to that data downstream, but the model itself sees everything.

This is a question of architecture, not trust. If you trust Google as a custodian, that's a separate question from whether the model needs to see your SIN to answer your question. Archevi takes the position that it doesn't: PII is stripped from the document before any query touches the model, then restored in the answer you see. The model literally cannot leak what it didn't see.

3. Hosting and Canadian law

Spark inherits Google's data residency story, which is excellent for Google's purposes but worth a second look for Canadian families. Documents live on US-based Google infrastructure, subject to US laws including the CLOUD Act. The CLOUD Act allows US authorities to compel disclosure of data held by US companies even when the data subject is Canadian.

For most search-bar queries, this is a non-issue. For a will, a tax slip, or a child's medical record, jurisdiction matters. PIPEDA-aligned Canadian hosting is the cleanest answer for the documents you'd never paste into a search bar — and the only honest answer if Canadian privacy law is part of your decision.

What Spark is right for — and what it isn't

Use Spark for what it's brilliant at: your personal inbox, your calendar, your Drive, the agentic flows where always-on awareness across your data is the whole point. Spark will only get more capable, and Google has the model depth to keep pushing it.

Use a purpose-built family document vault — Archevi or something like it — for the files your household shares. The ones where one upload should serve every family member. The ones where PII should be stripped before any model query. The ones you'd want under Canadian law. Spark wasn't designed for those, and there's no setting that turns it into the tool that was.

The bottom line

Gemini Spark is a powerful personal agent. "Personal" is in the design, not just the marketing. For family documents, you want family architecture: a shared vault, PII redacted at the model boundary, hosted under the law you live under. That's what Archevi was built for.