Moving Past Data Slop: Real Talk on Agentic Data Modeling w/ Kyle Nesbit (Credible Data)

In this episode of The Joe Reis Show, I sit down at my house in Salt Lake City with Kyle, founder of Credible Data. Before founding Credible, Kyle spent nearly 17 years at Google building core distributed systems—including the in-memory shuffle/join backplane that powers BigQuery—before moving to Gradient Ventures and spending four years helping integrate Looker into GCP. We dive deep into the reality of semantic layers, the origins of Malloy, and why the current hype around semantic layers often misses the mark. Kyle breaks down why SQL conflates data modeling and queries, how agents are shifting data work from painful bottom-up ETL to deterministic top-down domain modeling, and why giving an LLM an entire enterprise schema creates nothing but "data slop." We also get into the future of data engineering, what analysts should actually be learning today, and why unlocking the "long tail" of dark business data is where the true value lies. Website: ⁠credibledata.com⁠ Timestamps 00:00 – Welcome to SLC & cheap rental cars 00:46 – Kyle’s background: 17 years at Google, BigQuery backplane & Looker 01:31 – Evaluating Looker, the GCP acquisition & culture clash 05:10 – The origins of Malloy & open-source data modeling 06:15 – Launching Credible (and why MS2 had to go) 07:35 – The semantic layer resurgence: governance vs. accessibility 11:24 – "Big Data" is just data: Why Kyle hates buzzwords 12:28 – The semantic layer as a dimensional interface 14:38 – Semantic layers vs. data modeling: The transform function 16:04 – Are semantic layers dead? Reproducibility in the agent era 19:58 – How SQL conflates data modeling and querying 21:28 – What Credible does: Top-down business domains with agents 24:45 – Deterministic modeling: Closed-form Fourier transforms for business 27:14 – Why bottom-up data stacks piled up & how agents flip the script 30:26 – The "Data Slop" problem and being slop janitors 33:04 – Multi-agent systems, data mesh, and avoiding hallucinations 36:45 – Why typed systems (Malloy) prevent agent mistakes 37:37 – Credible as the "TypeScript of Data" 39:15 – What should we actually teach data analysts now? 42:00 – The irreplaceable 10%: Strategy, data literacy, and business context 43:24 – Creative workflows with AI & teaching kids 47:00 – How enterprise execs are reacting to AI adoption 50:35 – The long tail of dark data (and the real value of Excel) 56:15 – Wrapping up & where to find Credible