During an internal AI overhaul at Netflix, Mckenzie Lock encountered a funny problem. Her team built an algorithm that, unchecked, would have transformed the Netflix homepage into a sea of Julia Roberts portraits.

The algorithm did exactly what it was designed to do: generate in-app artwork that would drive up viewership for smaller titles (and featuring a still of someone in the cast proved very effective for this). But a homepage full of the same good-looking actor isn’t very good for the overall product experience.

“You can train a model to optimize for almost anything. It’s people who decide what’s worth optimizing for,” says Lock.

She spent seven years leading the product org that’s responsible for distributing everything Netflix produces, where she oversaw an initiative to bring AI into core creative and production workflows. The Julia Roberts problem was a reminder that any successful AI transformation still needs human judgment at the center.

Given creatives’ general apprehension around AI, adopting AI in entertainment is a bit more fraught than at the average tech company. But amid the debates over which harness to use or models to train, this existential question still looms large for everyone: How do you figure out what you should do without getting distracted by what you can do — especially when what’s possible with AI feels more temptingly expansive by the day?

In this essay, Lock offers tactical tips for how teams can get to their own “should” with an internal AI strategy — and get everyone onboard, even the skeptics. She shares:

• How to structure internal debates with folks on the frontlines of workflow changes
• Why you should get creative with how you pitch your future state with AI (like literally illustrating it)
• How to make a list of AI bets that have genuine, even uncomfortable tradeoffs