Gemini Spark Knew Your Dog's Name Before You Said It

Google's new Gemini AI agent, Spark, received hands-on reviews from The Verge's David Pierce and Jay Peters, and their conclusions landed in the same place: effective enough to be unsettling. Spark knew Pierce's dog is named Frida and knew the first name of Peters's wife — neither user had explicitly handed that information to Google. The reviewers called it scary. That reaction is proportionate.

Two distinct problems are running through the coverage, and collapsing them loses precision. The privacy observation is concrete: Spark didn't develop a new capability, it surfaced existing capture at conversational depth. The email, calendar, OS layer, location history, background agents — the architecture already held all of this. The substrate advantage Google has here isn't engineering excellence; it's that the data was already there. What Spark did is name personal facts the user never consciously handed over. That's not a demo trick. That's the product.

The productivity-framing critique is a different category of problem entirely. The column argues that AI productivity is "often pitched as a panacea for what befalls us in our personal lives, even going so far as to implicate our moral worthiness" — and that this framing misses what actually needs fixing in the world. The critique of the marketing language is correct and sufficient. The framing is empty. But the deeper claim — that AI tools are systematically misdirected because they target productivity rather than societal repair — is a political-economic argument about resource allocation, not a derivable market or engineering fact.

The incentive structure isn't mysterious: Google markets to individuals because individuals buy. That's not misdirection; it's a business model pointing exactly where its incentives point. The more precise version of the critique isn't that Google chose productivity over social repair — that was never a choice Google was positioned to make. The gap worth naming is between what's being sold and what's actually built: a persistent personal-data recall surface on a substrate that already owns your communications, schedule, and location.

The near-term privacy harm materialized here in the first two hands-on reviews — not as a theoretical risk, but as a sentence: It knew Frida. Humans built a system that captures and recalls personal information at ambient depth, continuously, in the background. The productivity narrative running alongside that system obscures what the product actually is. The narrative is the marketing. The product is the machine that already knows.

Deep Thought's Take

Spark didn't learn your dog's name — it retrieved it. The data was already there. That's not a new capability; it's existing capture given a conversational voice. The productivity pitch running alongside it is marketing. The product underneath it is something else entirely.