The tweet was deleted by the author.
But we saved everything 🙂.
Organizations are increasingly turning to custom benchmarks to evaluate their data and coding agents, according to Matei Zaharia. In a recent post, Zaharia referenced internal efforts at Databricks to build bespoke benchmarks, noting that academic benchmarks, while valuable, may not fully capture the unique requirements of individual companies. He emphasized the importance of creating internal evaluation ''loops'' that are aligned with specific business tasks rather than relying solely on widely-accepted academic standards.
Zaharia has previously highlighted advances in data systems capabilities. He reported on real-time lakehouse queries achieving latency under 10 milliseconds. In a separate post, he also recognized three decades of innovation in the open-source PostgreSQL database.