Intelligent Data

Intelligent Data is the Cycle capability that sets up and cleans up the data a test needs, authored with the help of an AI coding agent. The files it works from are called Blueprints, and that section covers what a Blueprint is, how to write one, where Blueprints live, and the ways to run one. This page covers the other half: how an agent takes on that authoring, and what stays with you.

Why author test data with an agent?

A Blueprint is declarative YAML with a strict schema, which is exactly the kind of artifact an agent handles well. The library around it is readable too, so an agent connected to the Cycle MCP Server can study how your team already works before it writes anything. The agent takes on the typing and the recall, and judgment about which records a test needs stays with you.

Authoring by hand has a few hidden costs:

• Referencing a schema with each change: Every entity, action, and field source has a required shape, so writing one unaided means holding that vocabulary in your head or keeping the reference open beside you.
• Convention drift: Starting from a neighboring Blueprint and editing it is how one library accidentally grows three different shapes for the same kind of record.
• Mistakes found late: A structural error in a hand-written file often announces itself only after a record has landed in a system other people share.
• Manual cleanup: Data left behind by an experiment is data a teammate trips over next week.

Working with an AI coding agent

A typical agent-assisted loop looks like this:

1. Describe the data state you need in plain language, such as an outbound order with two lines for distinct parts in a given warehouse.
2. The agent reads the library to learn the system's connections, existing scenarios, and conventions before writing anything.
3. The agent drafts the Blueprint and validates it structurally, which catches schema errors with no system access.
4. The agent runs it, against a live connection or offline against a local database, and reports the values it realized.
5. You review the result and the agent tears the data down when you are finished.

Validate after every edit

Ask the agent to validate each time it changes a Blueprint. Structural validation does not execute the creation or teardown, so it catches most authoring errors before any record is written.

Where Intelligent Data fits

Intelligent Data complements the other Cycle Intelligence capabilities. Discovery turns your documentation into test specifications and your answers into a recommended test plan, and Intelligent Data supplies the records those tests run against. Intelligent App Exploration then drives the live application so the test itself is written against real screens and locators.

For the Blueprints themselves, including what they are, why they beat hand-built data, and how the engine runs them, start with the Blueprints section.

Authoring your first Blueprint

• Set up the Cycle MCP Server so an agent can author Blueprints with you.
• Learn what Blueprints are and how a test consumes one.
• Build a Blueprint from scratch for the structure and the full field vocabulary.
• Validate and run a Blueprint from a test, the command line, or an agent, including offline smoke testing.
• Keep variables and placeholders open while authoring, for the keys a run publishes and how a {token} resolves.
• Study advanced techniques for API record resolution, validation rules, and library scaffolding.