Materials from the ‘Data Platform in 2026’ episode (#Data)
Materials from episode 28 of Research Insights Made Simple, released on 24 August, are ready. Together with Nikolay Golov and Alexander Filatov of Tengri Data, we went from the boundary between OLTP and OLAP to permissions, isolation, and validation of AI-agent queries.
We discussed:
- How to tell from a real workload when an operational DBMS can no longer handle analytics;
- Where a classic MPP warehouse reaches its limits and what changes when storage and compute are separated;
- What a Lakehouse on S3 and Iceberg consists of in practice: a catalog, compute engine, permissions, cache, small-file compaction, and the operation of distributed components;
- Why a migration from Vertica to Trino, Iceberg, and S3 takes years, and the long tail created by legacy computations and stored procedures;
- What a year of development, pilots, and sales taught Tengri Data: platform technology must be connected to a measurable business outcome;
- Why SQL access alone is not enough for an AI agent—it also needs a semantic layer, deterministic checks, strict permissions, and workload isolation.
All episode materials:
- Episode page
- Video: YouTube, VK Video
- Audio: Podster, Yandex Music, Apple Podcasts
- Text: Short transcript
If you still have questions after watching, leave them in the comments. I will collect them for a follow-up on data platforms—there is plenty more to explore here.
#Data #PlatformEngineering #Architecture #Database #AI4SDLC #Engineering