Problems discussed, let's pick one tomorrow 1. Given the thresholds we receive from apple around gaming how would you retroactively apply them the history of games in the user timeline to make that timeline more accurate. Assume they're perfect. Then once he gets going with that a. Assume they can be delayed up to 20m b. Assume they may or may not come from different shared devices and it's unknowable to us whether user has that setting on We like that one cus we all know it like the back of our hand so can focus on his problem solving and introduce new wrinkles as we go (of which there are endless but it's also a good thinker by itself even without those wrinkles). It loops in counterintuitive screen time BS which is an essential part of this role 2. Come up with a grammar for encoding rules that a parent might express to the agent in raw text so that those rules can be enforced deterministically • Consider blacklist and whitelist style rules • Consider budgets that may be daily, or weekly • Allow for differently weighted priorities of rules • Allow for ad-hoc changes (overrides) • What features of a rule would you explicitly disallow to preserve simplicity of the system? ◦ We've drawn this line at self referential rules like "make sure his time is 80% productive" Bit more of a pure algo /design question that doesn't roll in screen time quirks. We like this one because it's a real design problem that interfaces with product. There is a tradeoff between expressiveness and simplicity that we have to take a stance on and much of the elegance of the rest of our system rests on the choices we make with this grammar

CleanShot 2026-11-04 at 18.21.12@2x.png