Google’s latest Gemini Live announcement is built around a comforting message: “You shouldn’t have to guess whether a task requires Spark, a Daily Brief, or a quick inbox search.” The point, of course, is that the new voice-powered Gemini app can handle all three of those tasks automatically. But the sentence also reveals the exact problem with Google’s artificial intelligence strategy. People are being asked to learn the labels for different AI modes inside the Gemini app before they can feel comfortable using it. That is not a small friction point. It is a fundamental usability flaw, and it extends far beyond Google.
The Gemini app now invites users to move among Chat, Spark, and Daily Brief as if they were separate channels or even separate products. Each mode has its own icon, its own place in the navigation, and its own implied job. For a technology enthusiast, this kind of clear separation might feel tidy. For an ordinary user, though, it is a set of unnecessary decisions. Instead of simply telling the assistant what they need, they must first choose which area of the app is most likely to handle that kind of request. That is a lot of mental overhead for a product whose entire purpose is to remove mental overhead.
Engineering labels are not product features
There is a long tradition in software of naming internal systems and then letting those names leak into the public interface. Sometimes that works because the name is genuinely simple and memorable. More often, it leaves users confused about which tool is meant for which job. Gemini is a textbook example. Spark is not just a code name or an experimental toggle; it is advertised as an AI agent experience within the app. Daily Brief is less a separate product than a cleverly packaged version of what any good calendar assistant should offer by default. Yet both carry their own branding, and users are expected to navigate between them. Google may believe this gives the app a sense of richness and choice. The real consequence is that Gemini feels less like one assistant and more like a folder of demos.
Daily Brief finds ways to be both noisy and unsettling
Daily Brief is the most obvious example of a feature designed by engineers rather than by everyday users. On paper, it sounds useful: pull data from Gmail and Google Calendar, analyze it, and present a personalized agenda with proactive updates. In practice, Daily Brief fails to distinguish between an urgent task and a random suggestion. It may tell you to prepare for a meeting that is on your calendar, but it can also nudge you to follow up on something you mentioned in a chat, or resurface a Google search you performed days earlier. That is the opposite of a calm, helpful assistant. A notification about a research query that happened days ago does not feel relevant. It feels like the app has been watching too closely.
Consider an innocent sequence of actions: a user searches Google for information about college scholarships or looks up a local animal rescue. Later, Daily Brief sees those searches as signals and includes a reminder to continue that research. There is no meaningful sense in which the user asked for a follow-up. The data is being repurposed in a way that may cross a line personalization should not cross. The original goal of an AI-powered daily brief should be to get someone ready for the day, not to generate busywork out of old search history. Google, of all companies, should understand the difference between context and surveillance.
Spark would be better as a hidden capability
Spark is, in many ways, the opposite. It represents the more advanced side of Gemini: an agent that can take action, perhaps by making a reservation, composing a message, or automating a multi-step workflow. That is exactly the kind of capability that should feel magical to consumers. Yet Google has decided to market Spark as its own brand within the larger Gemini ecosystem. Users are asked to think about whether their request requires Spark or Chat, as if the assistant is still deciding what it is capable of. That undercuts the product’s usefulness.
In a more mature assistant, the underlying model would inspect a request and determine when agentic capabilities are needed. If someone asks Gemini to schedule a ride to the airport, the app should simply spin up the necessary internal agent, complete the task, and report back. The user should not have to press a separate button or move to a separate tab labelled Spark. Naming the agent mode gives it a separate identity, but it also gives users a new concept to learn. That is especially harmful for an AI product targeted at a mainstream audience.
The rest of AI is making the same mistake
Google is not the only technology company creating this problem. Anthropic’s Claude app now separates Chat from Cowork modes, and users have to decide which of those surfaces they need before getting started. Until recently, the two modes did not even share memory of previous conversations. OpenAI’s ChatGPT makes a similar argument with its Chat and Work offerings, requiring people to switch contexts to signal whether an answer is meant for general curiosity or for their job. In each case, an implementation detail that might make sense inside the company—where teams build different underlying technologies—has become a visible, public interface choice.
Consumers are not necessarily wrong to want separate contexts for personal and professional use. But asking people to switch modes by brand name is a shortcut for designers who are struggling to make the model behave differently for different tasks. A more thoughtful approach would infer context from the conversation, the user’s known preferences, or the apps being referenced. The user should not have to know that a given feature is powered by a separate tool or team. The user should just ask.
Apple’s understated Siri approach might win
Apple’s recent moves with Siri are sometimes described as anticlimactic, especially compared with the aggressive product launches coming out of Google and OpenAI. But a quieter strategy could have a major advantage. Apple does not require iPhone owners to learn a new app, a new mode, or a new brand. Instead, it has been embedding intelligence into the interfaces people already know: Spotlight search, the Photos app, camera features, and Siri itself. Users can simply keep using their devices the way they always have and let the improved models do more behind the scenes.
That may be the only strategy that truly scales to a large consumer base. Asking billions of people to adopt a new mental model of “agentic AI” is a big request. Asking them to type or speak naturally to a familiar digital assistant is not. Apple is not necessarily building the most impressive AI demos, but it is building an environment where the technology gets out of the way. In a market where every other company is shouting about models, modes, and brands, getting out of the way may prove to be the strongest feature of all.
The rise of text-first assistants
Another signal that consumers want simpler interfaces comes from the growing number of text-based AI services. Startups such as Poke, Ollie, Lindy, Orchid, Lucas, Folk, Tomo, and Instinct are building assistants that live almost entirely inside messaging. There is no dashboard to configure, no separate modes to choose, and no cognitive load associated with navigation. Users send a text, and the assistant responds or takes action.
Text messaging has become a universal user interface. It is fast, familiar, asynchronous, and available in the same place as the rest of a person’s relationships. As one investor in this space, Justine Moore, has observed, people do not want to open an app every time they need help. They want “a contact they can text like a friend,” and the gold standard for that experience is iMessage. The rise of text-first assistants is therefore an implicit bet that the biggest cost in AI adoption is not model intelligence but interface friction.
This is the standard against which Gemini’s many modes should be measured. A truly helpful assistant should not make a user decide between Spark and Chat, or between a daily brief and an inbox search. It should listen to the request, understand the user’s context, and determine the right tool for the job automatically. If Google cannot do that, then Spark, Daily Brief, and everything else will still feel like products looking for a user.
Source:TechCrunch News