Netflix put a new button in the top navigation of its TV app on November 3. It is called "Discover", it carries a red gem icon and a "New" label, and it does not lead to another curated row. It leads to a dedicated tab where the recommendation engine that made Netflix is asked to step aside.

The first experiment inside it is built around moods. Instead of scrolling a wall of titles, you pick a theme — "Something Light", "Scare Me", "Family Movie Night", "Quick Watches", plus sets for what's new, what's popular, or what you've been watching lately. Each theme is previewed with a clip from a show or film, which is the difference between a list of words and a shelf you can feel: you see what you are agreeing to before you choose it. Picking a theme opens matching recommendations, from which narrower themes surface until something fits. If nothing does, "Something Different" resets the board.

The mechanism that matters is a small button. Every title carries a "Tune Your Recommendations" control next to "Play" and "My List". Use it to keep a recommendation, or to tell the service never to suggest it again. Netflix remembers that inside AI Discover. It is a quiet departure from the logic of the thumbs-up: this is not a verdict on the title, it is a statement about the evening. "Not tonight" is a different signal from "bad", and the interface finally treats it that way.

Netflix describes the tab as powered by a new intelligence layer it has been building, one that responds to what members say in real time and keeps learning. The company says it explains your results along the way — unusual for a recommender, an industry that spent years building in the opposite direction, toward signals the user never sees.

The rollout is deliberately narrow. AI Discover starts in the US, for members using Netflix in English on supported TVs — not the app, not the web, not phones. More users follow over the coming weeks. Starting on the television is not incidental: that is where the endless-scroll problem hurts most, and where the result is easiest to judge as a product rather than a demo.

Netflix calls the tab a "test kitchen", where it tries out new recipes for discovery and lets usage and feedback decide which ones to keep, improve, or serve next. Some may make their way into the main Netflix experience. The phrasing is careful: AI Discover is not a promise, it is a place where features have to earn their way into the main product.

For a company whose recommender was once its defining technical advantage, that is a notable shift in posture. The old logic optimised for what plays next — a ranking over titles. The new logic asks first what someone is after, and only then what matches. That sounds like a small difference and is structurally a different one: a ranking improves by collecting more behaviour, while an intention has to be asked for.

The open question is whether people want to articulate a mood when they sit down to switch off. The answer will not arrive in a press release. It will arrive in the usage of a tab most members will not see for weeks.

Sources