I lead Firebase for Builders at Google. This site is where I take production AI failures apart in plain words and build the instruments that show you where yours will break.
I am a Developer Relations Manager at Google and the lead for Firebase for Builders. The work is mobile and full-stack AI enablement across the Google developer ecosystem: workshops, technical writing, and carrying what breaks in the field back to the product teams.
ulukaya.dev is the personal side of that work. I write about the places where production AI systems break, from the bill that arrives after an agent loop runs all night to the coding agent that passes every test and still ships the wrong change, and I show what fixed it. The essays take one failure apart at a time. The instruments let you run five of those failures in the browser. Thenotes are the short form: one gotcha, one fix, and the code that applies it. Theblueprints are the reference architectures the essays keep coming back to.
Every model call goes through a gateway that checks who is asking before it spends anything. On Firebase that is App Check attestation in front of Cloud Run, so scripted traffic never reaches the meter.
An agent loop with no ceiling is a memory leak that bills you. Budget the context, send cheap work to cheap models, and put a spend cap and a circuit breaker in front of the meter before adding concurrency.
Real systems are brownfield. Adding AI by rewriting a working backend produces regressions; adding it as a sidecar next to the existing service, then moving traffic over one route at a time, does not. I build and run my own work on the same infrastructure I recommend to builders: Firebase App Hosting, Cloud Run, and Firestore.
For talks, workshops, or a question about anything on this site, email is the fastest route.
Everything on ulukaya.dev, the essays, notes, instruments, and blueprints, is my own work and my own opinion. None of it is official Google policy, a binding SLA, or a commitment about unlaunched products.