Inside the Three-Layer AI War for the Living Room
By Fernando Amendola and Thierry Fautier
Three companies are building three architectures for the same asset: the viewer’s intent. Netflix wants its application to become the intelligence. Google wants the operating system to become the orchestrator. Operators are splitting over how much of the front door to hand a frontier model.
The television industry is undergoing one of the most important architectural transitions since the migration from broadcast to streaming. For a decade the consensus was simple: aggregate content, engineer a passable recommendation carousel, capture subscription revenue. That model is now obsolete.
AI is changing the locus of control. The next competitive layer is no longer the content catalog, but the conversational intelligence layer sitting between the viewer and discovery. The battleground has moved from the database to the dialogue. If you do not capture the user’s intent, your catalog becomes a silent, interchangeable commodity.
This is not theoretical. It is actively deploying. YouTube has rolled a conversational assistant into its TV app. Netflix has opened an OpenAI-powered search beta. Liberty Global went live with SuperSearch at Sunrise in Switzerland in 2025, and has since extended it to Belgium. At CES 2026, Google and Amazon both widened assistants already shipping on their platforms.
Tracking those deployments across the sector, The Media League, working with the AI Intelligence in M&E the AI² Research program, have mapped the ecosystem into three distinct strata of conversational intelligence. Each pursues a different strategic goal, optimizes for different margins, and demands an entirely different technical pipeline. Deciding where your organization anchors itself is the defining economic mandate for the next three years.
Layer 1 — Domain AI: The Netflix Sovereignty Play
Netflix is not merely improving search. It is attempting to transform the Netflix application itself into a conversational entertainment intelligence layer. The application becomes the Intelligence.
Instead of retrieving titles by genre or actor metadata, the system infers emotional intent, contextual viewing state, cognitive load, and pacing preference. Netflix controls viewing history, pause and rewind patterns, completion rates, abandonment signals and scene-level metadata.
Netflix’s largest AI advantage is not its model access. It is its proprietary semantic and behavioral graph. Unlike platform providers, Netflix never has to normalize inconsistent metadata schemas from third-party services. It owns the content intelligence layer and the behavioral intelligence layer simultaneously.
Read the roadmap, not the beta. The search Netflix has shipped so far returns universal results, driven by the prompt and title metadata rather than individual watch history. Netflix has stated the next phase connects that search to its personalization engine. Weigh that against the record: this is a company that has spent two decades shipping what it said it would ship. The behavioral graph is not a description of the current beta. It is the declared destination — and Netflix reaches its destinations.
The architecture is optimized for one objective: maximize engagement inside Netflix. If AI discovery migrates to the operating system layer, Netflix risks becoming a commoditized content supplier underneath a universal assistant. They are building a walled garden of intent.
Expect Disney+, Warner Bros. Discovery and Apple TV to follow. Amazon Prime has taken the opposite path, relying on Alexa+ ported into the Fire TV environment — which places it structurally in Layer 2.
Layer 2 — Platform AI: The Universal Orchestrator
Layer 2 belongs to the hardware and OS gatekeepers: Google TV, Apple TV, Roku, Fire TV. Their mandated role is universal orchestration and assistant ownership. Google’s architecture is the inverse of Netflix’s. Where Netflix wants AI embedded inside the application, Google wants AI above all applications. In the Google TV model, Gemini becomes the universal orchestration layer. The user does not interact primarily with Netflix, Disney+ or Prime Video. It interacts with Gemini.
Google’s core technical challenge is federating heterogeneous metadata schemas, entitlement systems and inconsistent taxonomies across dozens of providers — a classic federation problem demanding universal embeddings, metadata normalization and entitlement resolution.
Its advantage lies in Android TV distribution, cloud inference scale and household context. Unlike Netflix, Google can combine device telemetry, smart home context, cross-device behavior and persistent assistant memory — the foundation of an AI-native household assistant spanning television, commerce, search and advertising.
This puts Google in direct structural conflict with content providers. If Gemini controls conversational discovery, Netflix loses recommendation ownership, engagement loops weaken, and applications become interchangeable content sources underneath the platform. This is the hidden battle: who owns the conversational relationship with the viewer?
Layer 3 — Aggregation AI: The Pragmatic Operator
Liberty Global, Comcast, Sky and Canal+ represent a third, far more pragmatic architecture. They are not building an AI-native entertainment agent. They are not building a universal AI operating system. Their strategy is semantic aggregation.
Operators must unify live TV, streaming apps, catch-up, FAST channels and third-party ecosystems. They do not own the content graph. They do not own the operating system. But they own the entitlement layer — and that is not nothing.
The classic operator pipeline is deliberately conservative: it enhances the existing navigation layer rather than replacing it. Deterministic interfaces, low hallucination risk, predictable UX. Less disruptive. Also less transformational.
But the moves now on the board are not uniform. Canal+ will replace keyword search in its app with natural language from June 2026, powered by OpenAI, while running a separate Google Cloud partnership to index its library. JioStar has integrated OpenAI into JioHotstar as Multilingual Cognitive Search, with a two-way tie-in that surfaces its catalog inside ChatGPT itself. 3SS’s 3Ready Hero remains the closest off-the-shelf option for an operator that wants the capability without the commitment.
The aggregation layer is therefore splitting. Some operators enhance the interface and keep the model at arm’s length. Others hand it the front door and accept the dependency. That divergence, inside a single layer, is the fight worth watching.
Three Pipelines, One Contested Asset
Modeling the economic impact means comparing the pipelines directly. Let’s break them down as follows:
Netflix (Domain): Voice → Intent Parsing → Proprietary Engagement Graph → Recommendation Ranking → Generative Conversational Response
Google (Platform): Voice → Gemini Intent Parsing → Cross-Service Retrieval → Metadata Federation → Universal Ranking → Playback Invocation
Operator (Aggregation): Voice → Semantic Query Expansion → Metadata Federation → Cross-Service Retrieval → Traditional TV UI
Three architectures. Three cost structures. One contested asset.
The Economic Mandate
Television is becoming an intent interface. This is not an iterative UI update. It is a fundamental realignment of unit economics.
Netflix wants the application to become the AI. Google wants the operating system to become the orchestrator. Operators want AI to simplify the platform they already control. Three architectures, one contested asset: ownership of viewer intent. Whoever controls the conversational layer controls engagement, recommendation, advertising, personalization and the viewer relationship itself.
The action items are stark. Stop treating conversational AI as a novelty feature. Anchor your strategy on one question: do you have the proprietary data density to defend your own application layer, or must you architect your metadata to be ingested seamlessly by universal orchestrators? There is no third answer.
Decide which layer you intend to dominate, and dismantle any roadmap that does not advance that specific goal. 2025 was the year of first deployments. 2026 is the year of acceleration. If by 2027 you have not integrated this capability into your service, you will be playing catch-up — and catch-up in an intent war is not a position, it is a concession.
About the Authors
Fernando Amendola is Lead Marketing Strategist and co-founder of The Media League, and the AI Intelligence AI² Research program leader. A pragmatic changemaker with 30 years in Media & Entertainment, he aligns strategy, go-to-market execution and portfolio evolution to market trends and technology innovation.
Thierry Fautier is Lead Business Strategist and co-founder of The Media League. A Silicon Valley veteran and Technical Emmy Award winner, he played a key role in driving Ultra HD and real-time video compression technologies, and works across strategic partnerships, industry alliances and media technology innovation.
About The Media League
The Media League is an independent Media & Entertainment advisory practice co-founded by the authors, connecting operators, broadcasters and technology vendors with senior domain expertise across the content supply chain.
About the AI² Research Program
AI² Research is an independent AI Intelligence research program, tracking artificial intelligence deployment across Media & Entertainment. It maintains a forensic corpus of documented deployments structured against a seven-pillar taxonomy spanning production, distribution, experience and governance, drawing on subject-matter input from The Media League and a wider network of partners and advisors.