Google's Gemini Spark Runs All Night. Your Funnel Still Assumes Nobody's Home.

Every conversion mechanic your marketing team relies on assumes a human is sitting at a keyboard right now, paying attention, ready to act inside a narrow window. Live chat pop-ups. Fifteen-minute retargeting cookies. "Book a demo before you leave this page." Gemini Spark just broke that assumption, and almost nobody in marketing has noticed yet.

Google introduced Spark at I/O 2026 as a personal AI agent that runs continuously on cloud infrastructure, not on the user's device. It stays alive after the laptop closes. It monitors Gmail, manages a calendar, drafts documents, connects to third-party tools through MCP, and is already being positioned to make purchases on a person's behalf. It is not a chatbot waiting for the next prompt. It is a standing process that acts on a schedule nobody controls but the user, at hours nobody in your marketing org is staffed for.

That is the part worth sitting with. The buyer researching your product at 2 a.m. used to be a human, tired, half paying attention, likely to bounce. Increasingly it will be an agent, running on Google's infrastructure, with no fatigue, no session timeout, and no reason to leave your pricing page before it has extracted everything it needs.

The Session Was Never a Neutral Design Choice

Marketing built an entire discipline on top of one quiet assumption: engagement happens in a session, and a session means a human. Live chat routes to a rep because a human is expected to type back within seconds. Exit-intent popups fire because a human's cursor moving toward the browser chrome signals they are about to leave. Retargeting windows are sized in hours or days because that is roughly how long a human's buying intent stays warm before it cools.

None of that logic holds for an agent like Spark. It does not get impatient waiting for a chat rep. It does not have exit intent, because it does not experience boredom or distraction. It does not need to be retargeted within 30 days, because it can simply be asked to check back next quarter and it will, without anyone remembering to re-engage it. The mechanics marketing built to manage human attention are irrelevant to a process that has none of the constraints attention imposes.

This is not the same story as "AI shopping agents completing checkout" or "AI purchasing agents evaluating your B2B site," both of which assume an agent doing one bounded task and leaving. Spark is different in kind: it persists. It comes back. It accumulates context about a user's preferences and history across weeks, then acts on that accumulated context without a fresh session ever being opened by a human at all.

What Breaks First

Three parts of a typical B2B funnel fail quietly the moment a meaningful share of traffic is agents like Spark instead of humans.

Gated content is the first casualty. A form that trades an email address for a whitepaper assumes a human who wants to read it personally and will supply a real, checkable identity to get it. An agent filling that form on a user's behalf has no reason to hesitate, no privacy instinct to overcome, and no attention span to reward with a PDF. The gate stops filtering for intent and starts just filtering for agent capability.

Live chat is the second. Routing every chat session to a human rep on the assumption that the visitor is a person who wants a human conversation misreads what is increasingly present: a background process asking structured questions to fill a structured comparison it is building for its user. Reps end up having conversations that were never meant to be conversational.

Pricing and demo-booking flows are the third, and the most expensive to get wrong. If a "talk to sales" CTA is the only path to pricing information, an always-on agent simply cannot complete its task, and it will report back to its user that your product could not be evaluated. It will not wait around for a callback. It moves to the next vendor on the list, because moving on costs it nothing.

The Readiness Gap Nobody Has Scoped

Most marketing teams already ran an audit for one-off AI shopping and purchasing agents earlier this year. Spark requires a different audit, because the agent is not passing through once. It is coming back, on a schedule set by its user, potentially for months.

Build for the Process, Not Just the Person

The teams that get ahead of this will not do it by writing more content. They will do it by making the parts of the funnel that assume human patience, human hesitation, and human forgetfulness optional rather than load-bearing. Pricing becomes a page an agent can read instead of a call it has to book. Product specs become structured data an agent can parse instead of prose it has to interpret. Re-engagement becomes something the agent can trigger on its own rather than something a marketer has to catch within a shrinking cookie window.

Spark is a beta product from one vendor, and it will not be the only always-on agent running by next year. Treat that as the point, not a reason to wait. The funnel built for a human who is briefly, distractedly present is about to compete with a process that never leaves. Build for the one that stays.