Ben Rothschild

I'm excited to share a project I've been working on to scale AI-native analytics patterns at Meta. At Meta, we noticed that most product analytics questions come back to a handful of patterns. How active are our users? Are they coming back? Where do they drop off? What's driving the topline movements? Even so, teams kept rebuilding the same metrics, retention tables, funnels and growth accounting from scratch, and the near-duplicate logic made analysis difficult. Our fix was to define each common pattern once, as a parameterized library, and redesign the stack so agents and humans can produce and consume analytics reliably. We call these Analytics Libraries. They cover seven core analytics patterns and power over 9,000 production pipelines across Meta. A few things we learned along the way: 1. Agents are much better at filling in a typed spec than at writing SQL. When the library generates the SQL, the core logic can't be hallucinated. 2. Agents also do better calling a simple query API than writing queries by hand. The LLM picks the model, metric, and filters and a deterministic tool turns that into correct SQL. 3. Analysis recipes, the steps an analyst would follow for "why did this metric drop?", give more reliable answers than asking an agent free-form questions. 4. Standardization isn't free. It pays off for recurring, high-trust measurements, but forcing early exploration into rigid schemas slows teams down. The main takeaway for any analytics team: find your patterns, standardize them, and document them well. You can't automate a mess, and the teams doing this groundwork now are the ones whose agents will actually work later. I wrote up the full approach, with examples of the specs, the query tool, dashboards, and more of the downstream capabilities this architecture unlocks in the Analytics at Meta blog: medium.com/@AnalyticsAtMeta/…

A great night for @PeteButtigieg! Come meet Pete and support the campaign in Brooklyn March 5th! secure.peteforamerica.com/on…

My new favorite tldr

After looking at package-lock.json file I now believe it's a miracle that any node.js or react app runs

Every Building on Every Block: A Time Capsule of 1930s New York nyti.ms/2GKfGWS

Clear skies and good vibes. Sending @samrothschild8 off to college, Purdue class of 2022. 👨‍💻📚 @ Purdue University instagram.com/p/BmeK40ElaT0P… – at West Lafayette, IN

Was very impressed by all the sessions from #AWSSummit... here are some of my favorite quotes #AWSSummitChicago @awscloud

Excited to visit the @awscloud Summit in Chicago this week! #AWSSummit – at Chicago, IL

I think I understand why people don’t do Uber pool anymore – at Chicago, IL

🤔🤔🤔 Or you could work with the democrats to raise the minimum wage