Reliable data agents may need mostly fixed code, not more AI
MIA is a marketing agent built on a BigQuery data warehouse and a platform. It works with messy data such as channel spending, model outputs, and planner s that arrive as deeply nested data. The main lesson from shipping it is that reliability comes mostly from , not from the .
The acts as a layer that reads intent and explains results. The trustworthy parts are typed, repeatable, and tested. With messy real-world data, an agent can sound confident while being wrong: it may invent a column, guess the wrong join key, or make up a number when a query returns nothing.
Instead of putting the whole schema into the prompt, the system uses a that connects business ideas to real database fields, join paths, and allowed values. Words like “revenue” or “current spend” are resolved through that graph instead of being left to guesswork.
Key points
- MIA runs on a BigQuery warehouse and a platform.
- The handles intent parsing and result explanation, not the core data logic.
- Reliability comes from , typed, tested code.
- A maps business terms to real fields, join paths, and allowed values.
- Messy data agents can invent columns, guess join keys, or fabricate numbers if too much is left to the model.