Step 1: Linear regression Step 2: realize linear regression solves the problem well enough. Step 3: mourn the fact that you won't get to break out the fancy stuff.

Don't let the wins go to your head, or the losses go to your heart

An ounce of requirements is worth a pound of development

A/B tests should be evaluated at a cohort level. What's good for highly engaged users often differs from what's good for infrequent users.

Bad compromise is worse than no compromise

Make a wiki page with links to docs that people ask you about frequently

Many of the best systems are emergent rather than designed

death, taxes, and some engineer deciding to name their service "Hodor"

Hope is not a strategy

You are the average of the five podcasts you listen to most

Humanity will create AGI before it creates a printer that works

Simple solutions are REALLY hard to create

Implement what Facebook did 2y ago and you'll probably make a ton of money

If you can't say no, you can't prioritize

Feelings are good for you. Explore them.

Speed isn't the bottleneck on communication -- quality is

Skepticism when creating the schedule. Optimism when executing it.

Even when schedules slip, they still have value

The weird ML hill I will die on: AUC is a bad acronym

The hardest problem in computer science is opportunity sizing

Surely Equifax would never leak all of my data I never consented to give them