Research Insights Made Simple #28: the data platform in 2026—from DWH to Lakehouse and AI agents (#Data)

I almost forgot that today at 16:00 there will be a live stream about data platforms. It continues the topic of the sixth episode of Research Insights Made Simple, where Nikolay Golov and I explored how to build data platforms in 2025. In episode 28, we return with a harder question: what happens when a neat storage, compute, catalog, and orchestration design meets legacy systems, migration costs, and real analytics workloads?

Our guests are once again Nikolay Golov, joined by Alexander Filatov. Nikolay is Chief Product Officer at Tengri Data; previously, he led data platforms at Avito and ManyChat and taught at Harbour.Space. Alexander moved into DWH from backend development, having built data-processing tools in Python and C++. He spent nearly ten years developing Avito's data warehouse, worked on the migration from Vertica to Trino/Iceberg/S3 in 2022–2025, and now leads engineering at Tengri Data.

We will discuss:

• Where the boundary between OLTP and OLAP lies, why analytics on a replica quickly hits a ceiling, and why “let's just add ClickHouse” is not yet an architecture;
• How read and write patterns shape a system, and why engineers must distinguish what an analyst asks for from what they actually need;
• When a classic MPP warehouse becomes a bottleneck and what separating storage from compute in a Lakehouse delivers in practice;
• How to move to a new stack without stopping analytics: parallel systems, result validation, cost, and operational risk;
• How AI agents change platform requirements—from metadata and access control to action transparency and resource governance;
• Whether this ultimately requires a separate AI-native data platform, or whether the foundations remain the same.

I want to move beyond a catalog of fashionable technologies and examine the engineering trade-offs: when migration is genuinely necessary, what it will cost, and which decisions hold up not only in a presentation but also in production.

#Data #DataEngineering #Databases #Architecture #PlatformEngineering #AI