Run in your warehouse
We provide a lightweight dbt macro. Your data team drops it into your project and runs it against your Snowflake or equivalent warehouse. Compute stays in your environment.
Most enterprise AI pilots fail because the historical evidence does not support the automation. AGX starts with a standardized extract, tests whether the data is usable, and returns a clear Go/No-Go report before anyone funds a pilot.
AGX does not need read-only production credentials to begin. Your team runs the extractor directly inside your warehouse, and the engine runs the math.
We provide a lightweight dbt macro. Your data team drops it into your project and runs it against your Snowflake or equivalent warehouse. Compute stays in your environment.
The macro pulls the exact fields required by the AGX Evidence Ingestion Schema and anonymizes sensitive data using SQL functions before the data ever leaves your warehouse.
Your team exports the resulting standardized Parquet file and drops it in a secure, encrypted AGX bucket. The diagnostic starts the moment the file arrives.
Before you run the macro, define the domain grammar. AGX does not need your credentials; it needs to know the metric, the decision, and the hard boundary.
The AGX Evidence Extract Pack requires a modern data stack. This takes 60 seconds to configure.
The diagnostic is not a workshop. The data either supports a falsification test, needs assembly, or cannot support an AI pilot yet.
Do the raw records map cleanly to the required AGX evidence fields?
Do evidence timestamps strictly precede decision timestamps? If an approval is logged after a release, the control is untestable.
Are there enough historical cases to falsify a hypothesis without overfitting?
What is the theoretical financial impact of the operational drift visible in this dataset?
Delivered immediately after the data drop. No slide decks. No vague recommendations. Just a clear, evidence-based classification.
Your evidence is clean, temporally consistent, and voluminous enough for a falsification test.
Next step: Fund the paid Falsification Pilot.The data exists, but temporal ordering, outcome tracking, or specific evidence fields need assembly first.
Next step: Apply the exact dbt transformations and schema mappings provided in the report, then re-run.Historical logs do not capture the decision context or final outcome. The workflow is still operating on gut feel.
Next step: do not fund an AI pilot. Instrument the workflow first.The diagnostic exists to save time, budget, and bad AI rollouts.
No production credentials. Your team runs the dbt macro in your warehouse and drops the anonymized file.
AGX runs the readiness checks directly against the evidence. No three-week workshops.
If the report says the data is untestable, you avoid a pilot that would have guessed business logic and failed audit.