Podcast: Tiger Teams, Evals and Agents: The New AI Engineering Playbook

The emergence of AI engineering as a distinct discipline forces us to rethink our data pipeline architectures fundamentally. Tiger teams and evaluation frameworks aren't just organizational patterns—they're becoming operational necessities as we deploy agents that need real-time observability and quality gates. From a practical standpoint, this means data engineers can no longer treat AI systems as black boxes downstream; we need to build instrumentation and feedback loops directly into feature engineering and model serving layers. The shift toward open-source AI tooling, which Mastra exemplifies, suggests that proprietary lock-in around model deployment is weakening. For teams building modern data stacks, this translates to investing in evaluation infrastructure alongside your dbt and dbt-core workflows. The concrete takeaway: establish cross-functional data review processes now, similar to code review but for model inputs and outputs. This prevents the expensive discovery phase when production agents start making systematically biased decisions that corrupt your downstream analytics.