How a title
earns your trust

The thing I owned is what we call evidence. All the little pieces of information around a title that help you decide. Number one this week. The final season. The actor you love. Based on the book you read.

Every piece has to be true. We are handing you real reasons, so when you press play, you trust we were being straight with you.

That line under the art, "Spent 6 weeks in the Top 10," is a piece of evidence. So is the trophy on the poster, or the "New Episode" badge. Every one true, every one earning its place on the art.

When I got there it was a mess, and a very human one. For about ten years, team after team had bolted their own badge or line of text onto titles. The same title might say four different things in four different places, and the loudest message usually won, not the most important one.

The home page had turned bright red and shouty. We had a nickname for it, the NASCAR effect. A race car with so many sponsor stickers you cannot tell what car it is. When everything is shouting, you cannot hear any of it. That is a person we lost, at 9:40 at night, because we could not get out of our own way.

Think of the product as a sand castle, and everybody pours their own pile of sand in. Something inevitably slips. My job is to keep the castle standing while it keeps growing.

This was the hardest part of the job. Every single team wanted more. More badges, more messages, more of their thing on the title. And every one had a good reason. From where they sat, they were right.

I could not win that by saying no louder than them, and I could not win it by letting everyone pile their stuff next to each other. So instead of arguing, I forced conversations, and I brought receipts to every one of them.

At Uber I had designed scooter unlock to reveal each step just in time. Same instinct, grown up. At Uber it was about sequence. Here it is about attention, matching how much the evidence says to how much you are giving it in that moment.

On the home page, skimming, you get the single strongest, most personalized reason and nothing else. One thing, the best thing. Hover or focus on a title and you have told us you are a little curious, so the evidence earns the right to say a bit more. Click into the detail page, where you are weighing whether to press play, and that is where the fuller story lives, because now you are reading.

Deciding when to speak, and how much to say, based on what a person wants in the moment, is the whole job of an assistant too. Same design problem, one surface over.

A principle on a slide is cheap. I ran a test. The old way let the strongest few categories win outright, so a title often said the same kind of thing twice, two flavors of popular, and the second line wasted the slot. I tested a rule: the stack could take only one callout per theme before it had to move down to the next one. Suddenly the second and third things a title said each added something new.

Getting there was not one clean move. I would design a stripped-down version. We would drop it into that shrinking, video forward frame, look at it, and realize it was still too much. So we would cut again. Me, my content designer, and the engineers, arguing over single words and single pixels, until what was left was only the things that earned their place on that screen. That narrowing down is the actual craft.

This is the one I love most, and it is the least glamorous, which is very on brand for me. Availability evidence answers one simple question: when can I watch this?

For most of Netflix's history that question barely existed. Everything dropped at once. Then came things with a clock on them, live sports, weekly episodes, games, and suddenly "when" was a real product question. It reordered the questions I designed around. What is this. When can I watch this. How do I watch it. And only then, why should I care.

At the time all that timing information was jammed into the same badge system as everything else, right next to "number one this week" and "new season." But availability does a different job. A promotional badge can rotate in and out. Availability has to be there the moment it matters. Treated like a promo, it kept getting bumped by louder messages.

So I made the case to pull it into its own evidence type, with its own component, rules, and priority. Anything that affects whether you can watch, not out yet, expired, not on your plan, now had a dedicated place and could not get buried. It sits below the title on purpose: research found higher comprehension when availability is paired with title-level information.

By this point so much of the org was using the framework that it had to work without me. So I wrote the whole system down: the hierarchy, the categories, the priority rules, composition, surface differences, edge cases, and the reasoning behind them. I ran office hours. I built short courses so designers, PMs, engineers, and partner teams could learn it. And I put a chatbot on top of the documentation to answer routine questions. Then it was called NotebookLM; now it is called Gemini Notebook.

The point was to let the framework scale without turning me into its permanent help desk.

Because the system had gotten that good, the one thing it still could not do started to bother me. It was still talking to everybody the same way. Netflix knew an extraordinary amount about what each member liked, and almost none of it ever made it onto the screen.

So the next question became: can the evidence tell you why this title might be right for you? That is what pulled me toward AI.