AWS Virtual Engineering Workbench cuts automotive test case creation time by up to 80% using generative AI
AWS built a generative AI extension of its Virtual Engineering Workbench that automates test case creation from automotive software requirements — reducing test case creation time by up to 80% while maintaining a human-in-the-loop review step. In a real-world deployment with Schaeffler, the system cut an experienced test engineer's preparation time from 820 hours to 265 hours for 837 requirements.
Background
Automotive software complexity has grown to a point where manual test case creation is a significant bottleneck. For a single complex electronic control unit with tens of thousands of requirements, and a ratio of three to five test cases per requirement, the manual effort reaches into the hundreds of thousands of hours for a full vehicle program. AWS Professional Services built the VEW AI extension specifically to address this bottleneck, targeting the classify-and-generate steps that consume the most engineering time.
What Was Implemented
- Generative AI extension integrated into the Virtual Engineering Workbench (VEW) cloud framework
- Four-step automated workflow: requirements upload → AI classification of requirement type → AI generation of test conditions → AI derivation of test cases
- Built on Amazon Bedrock with Anthropic Claude (Instant for classification, 2.0 for test case generation)
- Human-in-the-loop review and approval at each step before progression
- LiteLLM-based AI gateway for multi-model access, cost tracking, and rate limiting
- Session state preservation so engineers can pause and resume generation jobs
- Collaborative sessions allowing multiple team members to work on a single test generation run
- Deployed to Schaeffler in production within four weeks; architecture uses API Gateway, Lambda, S3, and DynamoDB
Results
The general VEW capability reduces test case creation time by up to 80% , per the January 2025 AWS blog. In the Schaeffler deployment specifically, the system accelerated test case generation by up to 60% : an experienced test engineer now prepares 837 system requirements' worth of test cases in 265 hours rather than 820 hours (reduced from ~1.02 hours per test case to ~0.32 hours). The solution was implemented in production in four weeks. The human-in-the-loop design also reduces the risk of system requirements remaining unvalidated due to human error.
Lessons
- A human-in-the-loop design is essential in safety-critical automotive testing — it preserves AI efficiency while maintaining engineer accountability
- Prompt engineering (role-playing as a test engineer, structured output formatting, precision instructions) is a critical factor in output quality
- Session state preservation and collaborative workflow support are required for production adoption at scale
- Even with strong AI assistance, an initial context investment (prompt templates, classification criteria) is required before benefits materialize