At dbt Summit 2026, Riot Games senior data engineer Jess Zhang described how her team used AI agents to rebuild a production data warehouse to engineering standards during a real Wild Rift event.
The old setup buried business logic inside large PySpark jobs that mixed orchestration, configuration and transformation code. Zhang named three costs: an "unboxing tax" to understand any single metric, a "testing tax" from inconsistent validation, and an "exclusivity tax" that kept domain experts out unless they learned the whole tooling stack.
The team piloted the agentic approach on Echoes on the Rift, a nostalgia event launched August 7, 2026 that linked players' mobile Wild Rift and PC League of Legends accounts across five years of history and more than 60 cross-title metrics. A single League of Legends PC event table held about 18 terabytes of data, roughly 129 billion raw game events that had to be combed through just to calculate basic battle stats like total kills and damage dealt.
The system layers context: an engineer-built system prompt defines the agent's persona and tools, a jointly built "playbook" layer encodes naming conventions and schema contracts, and a task prompt lets an analyst describe a need in one sentence. That context feeds an assembly line of five stations that discover source columns, profile their value ranges, author a silver model with standard naming and join patterns, build a gold-layer model, and validate the result before a human reviews the pull request.
Generated models are checked at three levels: syntax, compilation and execution, with dbt v2 running compilation-level type checks against the catalog in seconds instead of requiring a full warehouse run for every small error.
Benchmarked against a similar event Zhang built by hand the previous year, which took two months of protected, dedicated focus time, Echoes on the Rift took about two and a half months even though Zhang stayed on the on-call rotation, absorbed shifting scope and built the agent system from scratch in parallel.
Zhang argued the real shift is in the cost curve: previously the 200th model cost about as much engineering time as the first, while the agentic model pays the upfront cost of the playbook and evaluation system once and drops the marginal cost of each additional model toward token and compute cost. Engineers remain in the loop, she stressed, with their effort moving from boilerplate SQL toward data modeling precision, review, and separating SQL mechanics from business semantics.
In Q&A, Zhang acknowledged that preventing contributors from building redundant models is still an open problem, managed today through PR review and instructions telling the agent to check for existing models before building new ones.
Source: dbt Labs Blog