1. How do I use AI Agent Studio?
  2. Overview of ATLAS

Overview of ATLAS

Agentic Testing and Lifecycle Automation Suite (ATLAS) helps you create and maintain repeatable tests for Oracle AI Agent Studio workflows and agentic applications.

Describe the test scenario and expected behavior to your coding assistant. The assistant uses ATLAS to prepare tests, record or generate test data, run checks, evaluate responses, and generate reports.

You can use ATLAS to validate:
  • Workflow routes and expected outcomes
  • Fixed responses and structured outputs
  • Model-generated answers, summaries, and recommendations
  • Application panels and their underlying workflows
  • Multiturn conversations and supported human interactions
  • Model quality, performance, and AI-unit consumption

How Replay Works

ATLAS follows a record once, replay anywhere approach. During file replay, ATLAS provides recorded responses for eligible data-producing nodes and workflow boundaries. The remaining workflow components, including routing, code, transformations, and LLM or agent nodes, continue to run.

This approach lets you evaluate workflow behavior using known data. File replay still requires access to a compatible runtime and the models used by the workflow.

When to Use ATLAS

ATLAS supports different testing approaches based on what you want to validate. Use the following guidelines to select the appropriate testing approach for your workflows and applications:
  • Verify a fixed response and the route that produces it: File replay with output and path assertions
  • Evaluate a summary or other answer whose wording can vary: File replay with path assertions and semantic judging
  • Validate current data, authentication, or connected services: A focused live test
  • Test application startup or another application stage: An application-facing workflow test with the appropriate context
  • Validate a top-level application panel: An application panel test, supported by backing-workflow tests
  • Validate behavior across several interactions: A conversation test
  • Compare model configurations: Run the same tests and fixtures with isolated baseline and candidate labels
  • Monitor consumption against an approved allocation: Economic regression testing with an accepted economic profile

Key Terms

Here are some key terms related to ATLAS:
  • Replay fixture: Recorded or generated data stored with a test and used in place of selected live responses
  • Deterministic assertion: A check with a definite expected result, such as a required field, fixed response, or operation path
  • Semantic judging: Evaluation of a response against quality criteria when its wording can vary
  • Test synchronization: Identification and completion of test changes needed to keep generated coverage aligned with a workflow or application
  • Economic profile: A source-controlled definition of representative outcomes, expected production volume, approved AI-unit allocation, baselines, and economic evaluation settings