Google's pitch for the new Gemini app reads like a quiet reframing of what a brand even is. The update introduces Gemini Spark, described as "an agent to help you get things done around the clock," alongside a proactive Daily Brief. The promise is not a faster chat. It is a presence. Something that operates beside the user, under the user's direction, while the user is doing other things.

Set that against the model marketing teams still run on. A brand has a website. An app. A campaign landing page. A chatbot in the corner. Customers arrive, sessions begin, sessions end, dashboards count the visits. The whole apparatus assumes the customer comes to you, performs a task inside your surface, then leaves. Every metric, every team structure, every quarterly review assumes that round trip.

The shift Gemini Spark packages is not cosmetic. It treats the agent as an objective-pursuing process that runs continuously, not a tool the user opens. Google reports more than 900 million monthly users across 230 countries already moving through that surface. When that scale of consumer behavior normalizes always-on assistants, the destination model of brand interaction starts to look like a fax machine with a logo.

"Gemini Spark, an agent to help you get things done around the clock."

The interface is becoming embeddable infrastructure

If this were only a Google consumer story, marketing leaders could file it under "watch list." The parallel signal from the enterprise side is what makes it operational. Kaltura just shipped Agentic AI Conversational Avatars, real-time multimodal interfaces that combine speech, screen comprehension, and organizational context. More importantly, they shipped a Conversational Agents SDK that lets developers and coding agents "add real-time, avatar-based agentic AI conversations directly into their company's applications and digital experiences, integrated with their workflows, data, systems, and agentic engines."

The underlying knowledge layer is already running. Kaltura Genies are described as conversational agents that answer using only an organization's verified content. The interface, in other words, can move from text to live avatar without changing the grounding model. The same verified knowledge can speak through a help center, a product page, a video stream, an embedded widget, or a synthetic avatar inside a partner application.

That is what makes this a marketing infrastructure shift rather than a feature launch. The cost of building a persistent brand presence has collapsed. The components are commodity: a grounded knowledge layer, a conversational engine, a multimodal surface, an SDK to drop it into anything. Any company with verified content and a workflow can ship an agent that operates wherever the customer already is, instead of waiting for them to load a URL.

Marketing teams are organized for the wrong job

This is where the interpretive pressure lands. Almost every marketing function is built around attracting attention to a destination and counting what happens inside it. Sessions, scroll depth, page completion, time on site, click-through, qualified visits. The agency model, the martech stack, the attribution debates, even the org chart all assume the brand is a place.

A persistent agent inverts that. The job stops being "get them here" and becomes "complete their objective wherever they are." The customer states an intent. The agent routes it across products, partners, content, and transactions on the customer's behalf. Success is no longer measured by whether the customer arrived. It is measured by whether the objective got fulfilled, how cleanly, with how much override.

The risk for marketing leaders is treating Gemini Spark and its enterprise cousins as another channel to push messages into. That misreads the surface. A proactive agent under user direction is not a media buy. It is a contract. The user grants standing authority to pursue goals on their behalf. The brand's only way to participate is to operate inside that authority well enough to keep it.

A new measurement stack, built around completed intent

Once presence becomes the interface, the metrics have to change. Engagement is the wrong currency. A persistent agent that drives engagement is intrusive software. A persistent agent that completes objectives is infrastructure.

Four numbers start to matter more than the old funnel:

  • Objective completion rate. What percentage of stated user intents the agent actually fulfills end to end, without dropping the user back into a manual workflow.
  • Time to resolution. How long from intent declared to objective satisfied, including the agent's own routing, verification, and escalation steps.
  • User-override frequency. How often the user corrects, halts, or replaces the agent's decision. High override is not failure; it is the calibration signal telling you where judgment is thin.
  • Grounded-answer rate. What share of agent actions trace cleanly to verified content or sanctioned tools, versus extrapolation. Kaltura's Genies frame this as a feature; for brand agents, it is the floor.

None of these live naturally in a campaign dashboard. They live in operating telemetry. That is the architectural change. Marketing becomes a continuous engineering function with a measurement stack closer to platform reliability than to media planning.

Authority boundaries are the design problem

The technology to put a brand agent into a customer's daily life is now cheap. The question is whether the customer keeps letting it stay. That answer is decided at the design layer, not the model layer.

A persistent agent earns its right to operate when it acts only on claims it can verify and decisions the user can override. Standing authority should be narrow, scoped to specific intents, and revocable in a single gesture. Every action should leave a trace the user can read. Anything the agent cannot ground, it should escalate rather than guess. This is not a values posture. It is the operating constraint that keeps the trust contract intact long enough to compound.

This connects directly to two things we've covered before. The argument that agents are deployable but trust is what compounds the advantage sharpens when the agent is no longer a session tool but a continuous presence, because every action either spends or deposits trust. And the case that the agent productivity gap is a measurement problem becomes literal here: without objective-completion telemetry, a brand cannot tell whether its persistent presence is working or quietly burning the relationship.

What this actually opens for builders

The new design space is wide. A retailer can run an agent that watches for relevant restocks, price moves, and warranty events, then surfaces them inside a customer's calendar or messaging app instead of waiting for a site visit. A SaaS vendor can embed a Genie-style agent inside its product that answers questions, executes configurations, drafts comms, and routes requests across the customer's own stack. A media brand can ship an avatar that holds a continuous conversation across episodes, remembering what the user cares about and surfacing the next segment when context fits. A B2B company can publish an SDK agent that integrates directly into a partner's workflow, where the brand experience is the action the partner's user takes, not a separate visit.

None of this requires a new model. The capability is shipped. What it requires is rebuilding the marketing operating system around continuous intent fulfillment, with measurement and authority designed in from the start. That is closer to product engineering than to campaign planning, which is exactly the redesign most teams have been deferring.

The firms that move first will not look like they are doing better marketing. They will look like they have entered a category their competitors cannot reach without rebuilding their stack. By the time the laggards rewrite their measurement systems, the persistent presence has already become the relationship.

A brand that still makes you come to it is already a legacy interface.

Sources