Why Samsung Built a Phone With Three AI Assistants

Samsung's Galaxy S26 ships with three AI assistants. Bixby handles device controls and Samsung-specific hardware shortcuts. Gemini is the default general-purpose reasoning and search layer. And now, by saying "Hey Plex," users can invoke Perplexity for AI-powered web search. Three assistants, one device, zero consensus on which one wins.

This is not accidental. Samsung Galaxy AI is explicitly positioning itself as a "multi-agent ecosystem," and that framing signals something meaningful about where the AI assistant market is heading.

The One-AI Assumption Is Breaking Down

For the past two years, the implicit assumption in AI competitive analysis was that this space would consolidate: one assistant would win, the way Google won web search and Google Maps won navigation. The platform logic seemed to support it - whoever got embedded in the operating system would accumulate usage at scale, build better models, and lock in position.

That assumption looks less stable now. The problem is that no AI assistant does everything well. Gemini reasons effectively over long context and integrates with Google Workspace tools. Perplexity is measurably faster at synthesizing current web information - it's built around real-time indexing in a way Gemini's search integration has not fully matched. As we covered in How Perplexity AI Is Changing Web Search, the product is genuinely differentiated by retrieval speed and source transparency. Bixby knows Samsung's own hardware: shortcuts, settings, device-specific features that a third-party AI cannot replicate.

When the use cases don't overlap, there's no winner. There's a routing problem.

Samsung's three-AI configuration is a practical answer to that routing problem: run the assistant that's actually best for the task, let users choose, and don't bet the platform on one model.

What Perplexity Gets From This Deal

The distribution math here is significant. Samsung ships somewhere around 200 to 250 million smartphones per year. The Galaxy S-series represents the premium end of that volume - devices where users pay for capabilities and pay attention to what ships. Perplexity just secured voice invocation rights on a substantial share of Android flagships.

The comparison to search distribution is worth making directly. Google reportedly paid Apple $20 billion in 2022 to remain the default search engine in iOS Safari. The logic: search happens at the moment of intent, before the user opens another app. "Hey Plex" puts Perplexity in a structurally similar position for AI search queries. The invocation is hardware-level, triggered by voice before the user reaches for a browser.

That's not a guarantee of dominance. Habits form at the point of first invocation, though, and Perplexity just secured first-invocation rights for a lot of people asking real-time web questions on Samsung phones. That's a different category of distribution than being available in the App Store.

Embedded AI Is the Actual Trend

This is the second major announcement in recent months confirming that AI assistants are migrating from standalone apps into OS-level integrations. When Apple confirmed that ChatGPT, Claude, and Gemini would all ship within CarPlay - analyzed in What ChatGPT and Claude in CarPlay Tell Us About AI's Next Phase - the signal was the same: the interesting distribution action is not in the App Store, it's in the hardware layer.

Standalone AI apps require users to remember to open them. Embedded AI triggers at the natural moment of intent: you're driving, you're looking up a restaurant, you're trying to turn off notifications without unlocking your phone. The friction reduction is not incremental - it's the difference between a tool you consult and a tool that's simply there.

The business logic follows. Embedded distribution produces usage volume that a standalone app can't generate, which builds behavioral data and model feedback loops that matter for product quality over time.

What This Means for AI Companies Without a Hardware Partner

The companies watching this with the most at stake are the ones that have not secured hardware-level integration. ChatGPT is available as a Samsung keyboard plugin but lacks voice invocation. Anthropic's Claude remains largely app-dependent on Android. Microsoft Copilot has Windows and Surface integration, but mobile is still an App Store story for most users.

Earlier this week, a Google Cloud VP warned publicly that AI startups lacking genuine differentiation face an existential threat as foundation models improve and margins compress. The hardware distribution problem is adjacent: even a well-differentiated product may find its growth ceiling lower than expected if habit formation is happening around assistants that got to OS-level integration first.

That's not a death sentence for app-only AI products. Perplexity itself is proof that a clear use case can earn a hardware partnership. But it does reframe the strategic priority. Model quality is the price of admission. Distribution is the actual competition.

The Routing Problem Coming Next

There's an honest caveat to the multi-AI phone model that Samsung's announcement doesn't fully address. Three assistants means three trust models, three privacy policies, and three sets of capabilities users have to learn separately. For most people, choosing which AI to invoke for which task is cognitive overhead they didn't ask for.

The current Samsung implementation is explicit: say "Hey Bixby," "Hey Google," or "Hey Plex" and you get the respective assistant. Functional, but not the seamless experience the marketing implies.

The next iteration - which Samsung will likely ship in some form within a product cycle or two, and which Apple is almost certainly prototyping - is intelligent automatic routing. You ask a question, the platform decides which AI handles it. At that point, model providers are not competing for user preference. They're competing for placement in someone else's routing algorithm.

That's a fundamentally different competitive environment than the one people imagined when standalone chatbot apps first launched. Samsung just made it more concrete.