AI and Data Platforms: Research Insights Made Simple №37 Materials (#Data)
Materials for Research Insights Made Simple episode #37 are now available. On October 12, 2026, I spoke with Nikolay Golov about connecting AI to all of an organization’s data and the engineering work hidden behind the promise of a “unified platform.” Giving an agent access to several databases does not yet guarantee a consistent answer. In one place, data arrives late; in another, a large JOIN shuttles it between systems; elsewhere, convenience ends when the cloud bill arrives.
We discussed:
- Four approaches from Nikolay’s diagram: a conventional warehouse with prepared data marts, federated queries, a universal DBMS, and shared storage for different engines.
- What remains behind the zero-ETL promise: copies, delays, recovery after failures, and deciding which tables still need to move.
- How Microsoft, AWS, ClickHouse, and Google combine these approaches, and the trade-offs of combining transactions and analytics in Oracle and SAP HANA.
- Where Databricks and Snowflake are heading as they bring analytics together with PostgreSQL. In Nikolay’s assessment, the components are still being assembled into a coherent system; it is too early to call the problem solved.
- Why an open format does not remove operational work: frequent small writes to Iceberg, large numbers of files, and object storage limitations do not disappear.
Materials:
- Episode page with timestamps
- Video: YouTube, VK Video.
- Audio: Podster, Yandex Music, Apple Podcasts
- Written recap of the conversation.
If you are already connecting agents to corporate data, tell us where you hit a limit: access to sources, the meaning of metrics, freshness, or query costs?
#Data #DataEngineering #Databases #Architecture #PlatformEngineering #AI