Prophecy puts a visual review gate between a data agent and production workflows

Prophecy has expanded its AI data-preparation agents to Snowflake and released a review flow aimed at business users who cannot—or should not have to—audit generated SQL or Spark code. The September 8 release makes a useful product bet: for self-service data agents, trust may depend less on exposing code than on showing the resulting workflow and data before a user approves a change. (Prophecy)

The agent produces an inspectable workflow

A user describes a data goal in natural language; Prophecy says its specialized agent combines that prompt with metadata-based knowledge graphs to prepare data, map it into a target format and run analysis. The output is a visual workflow rather than only an answer or an opaque block of generated code. Users can inspect and refine that workflow, generate charts on the same canvas and deploy it on their existing cloud data platform. (Prophecy)

The release adds a step-by-step review interface for changes to workflows and charts. Prophecy says users can see what the agent added or changed alongside the produced data before approving the output. Projects also open directly into an analysis view after a question, while the underlying pipeline remains one click away. That separates the default business-user experience from the implementation detail without removing the implementation from inspection. (Prophecy)

Snowflake becomes an execution boundary

Prophecy already supported Databricks; the new release extends the agent products to Snowflake. The company says Snowflake customers can run self-service workflows without moving data outside their existing environment, while retaining the governance and access controls managed by the platform team. Its broader platform description also names BigQuery, but this announcement’s expansion is specifically Snowflake. (Prophecy)

For operators, the important distinction is between reviewable and safe. A visual diff can help a domain expert catch an incorrect join, filter or chart transformation, but the announcement does not describe automated policy checks, rollback semantics, approval roles or tests for whether reviewers reliably detect mistakes. It also publishes no accuracy, latency or production-adoption measurements. Those controls and measurements should be evaluation questions, not inferred product features.

Availability without a public price

The tools are available now in Prophecy’s Professional and Enterprise editions, according to the company. The release invites users to start free but does not give free-tier limits or paid pricing. (Prophecy)

The practical test is straightforward: give the agent a small set of representative preparation tasks, then measure whether analysts can identify seeded mistakes from the visual review alone. Prophecy has exposed the right control surface for that experiment. It has not yet published evidence that the control catches the failures enterprises care about.