Summary: TechCrunch’s May 30 hands-on calls Gemini Spark, Google’s cloud-based 24/7 assistant launched at I/O 2026, “actually pretty useful.” Six assigned tasks show strengths in email summaries, local events, and context-aware lists, but weaknesses in exports, links, and instruction detail. The reviewer says Spark remains largely limited to Google’s ecosystem until MCP support and more export paths arrive.
TechCrunch’s May 30 piece is a hands-on test of Gemini Spark, the 24/7 agentic assistant Google rolled out at I/O 2026 in mid-May. The assistant’s assignment is helping users “navigate your digital life”: clearing online to-dos, turning unread material into summaries, and keeping things such as expense spreadsheets in order. Because it operates on cloud virtual machines, a long task never depends on your device being present; Sundar Pichai’s launch quip made the point — yes, you can close your laptop.
The verdict comes with caution but lands positive — the reviewer’s own words are “actually pretty useful.” Just as clear is where the boundaries sit and where the identity questions start.
The Product Behind the Name
Gemini Spark was unveiled on May 19, 2026 at Google I/O as a 24/7 agentic assistant built deeply into Gmail, Calendar, Docs, Sheets, and Slides. TechCrunch’s reviewer reached it through early access; the piece stays silent on pricing and tiers.
The fundamental difference between Spark and other agent tools comes down to where it lives. Its tasks execute on cloud VMs and keep running server-side, while agent frameworks such as OpenClaw depend on the local machine staying awake. A Spark user can shut the laptop and the work goes on. Trading infrastructure for experience is a move Google knows well.
Six Assignments, an Uneven Scorecard
The reviewer handed Spark six kinds of work, and the scorecard came back uneven:
- Drugstore shopping research: it assembled Walgreens sale items, coupons ready to clip, and a plan for stacking them; one promo code turned out dead, but it compensated with buy-one-get-one offers and member rewards
- Day-trip packing list: it checked the weather and the event’s details on its own, suggested chairs, sunscreen, and water, and noted dogs were barred; the list cannot be exported to Keep — Docs or email are the only routes
- Teen summer activities: it returned a local list annotated with driving distances, yet never volunteered costs or dates
- Friday newsletter digest: it picked the top articles, summarized them, and attached links; each link passes through a Google redirect for one extra click, and a request for 5 items came back as 4
- Weekend-events roundup: merging web search with the local newsletters sitting in Gmail, it even dredged up an obscure local “Beaver Queen Pageant”; adding anything to the calendar takes a reply plus one more confirming click
- Price-drop tracking: it read the request as a check every two weeks — a cadence the reviewer finds too slow
Taken together: recurring email and newsletter summaries, local event discovery, and context-sensitive list assembly are its dependable strengths. Cross-product export, the link experience, and the finer points of reading instructions are where trouble keeps showing up.
The details carry real weight. The packing list went past the obvious, adding blankets, a light layer, a shopping bag, and an umbrella, and it drew the no-dogs rule straight from the event page without prompting. Contextual inference of that kind is what separates an agent from a search box padded with extra steps. Yet the same system answered a request for five newsletter items with four — precisely the sort of lapse that chips away at trust in a tool designed to run unattended.
The Lid Stays Shut, the Work Goes On: Paying for the Cloud
Cloud-residency doubles as the pitch and the limit. Round-the-clock execution comes with Spark being, today, helpless beyond Google’s ecosystem — Resy bookings are out of reach, and proactive flight-deal hunting is not on offer. MCP integrations, the review notes, are in the works; that is the step that would carry it from “Google’s in-house butler” to a genuinely general-purpose assistant.
The absent Keep support is revealing. An assistant that builds lists yet cannot file one into Google’s own note-taking app implies its integrations were arranged around the Workspace productivity core — mail, calendar, documents, spreadsheets — rather than the full consumer stack.
Over on the iPhone, Spark shows up as a separate toggle inside the Gemini app, and the hardware Activity Button cannot be mapped to it. The reviewer would also rather text the assistant directly than open the app every time.
A Brand That Never Needed to Exist?
The review lands its hardest blow on positioning: Spark had no reason to become a standalone brand. AI product names are already in oversupply, and one more only deepens the cognitive load while adding nothing. The reviewer’s argument is that these capabilities should be folded into Gemini itself, sparing users the mental split between “questions go to Gemini” and “tasks go to Spark.”
For product teams, the review earns its keep by pricing the real exchange rate of an agentic assistant: whether tasks get done tracks integration depth and how well the details are polished, not the raw capability of the model. The road ahead is easy to name — MCP support, plus more export paths. Until then, Spark remains a capable butler kept inside Google’s fence.