5 Things Every AI Engineer Should Know About Agent Sandboxes Co-authored with Ryan Ismert, published by Google Cloud Tech on X, September 1, 2026
We spent several quarters benchmarking, breaking, and running agent sandbox runtimes across Google Cloud and open-source stacks. The five takeaways:
Cold-start marketing numbers are measured on empty loops, not real runtimes (our real-world bootstrap: ~610 ms vs. a claimed 150 ms). The isolation spectrum is defined by what code you share with your neighbor.
Introduction to Agents Lead author, with Antonio Gulli, Shubham Saboo, Michael Zimmermann, and Vladimir Vuskovic, on Kaggle, November 2025
Day 1 of the Kaggle × Google 5-Day AI Agents Intensive, which reached 1.5M+ learners. It is written for developers, architects, and product leaders moving from proof-of-concept to production-grade agents:
Taxonomy: Level 0 (core reasoning) → Level 4 (self-evolving systems) Architecture: Model (“brain”), Tools (“hands”), Orchestration (“nervous system”) Agent Ops: LM-as-judge quality, metrics-driven go/no-go, OpenTelemetry traces, human feedback Security: the utility vs.