Designed to automate one of the most critical and work-intensive steps in the validation process
As vehicle platforms become increasingly software-defined, engineering teams are tasked with managing an ever-growing list of requirements, engineering data and system specifications, all while ensuring full consistency and traceability
With these demands in mind, automotive technology supplier Marelli, together with Amazon Web Services (AWS), has developed a next-generation, AI-driven System Test Generation (STG) agent. The STG agent is designed to automate one of the most critical and work-intensive steps in the validation process: the generation of system test cases from engineering system requirements.
“The STG agent represents an important step forward in how we validate solutions for software-defined vehicles,” says Daniele Russo, head of system performance optimisation in Marelli’s electronics engineering team. “By combining our engineering expertise with advanced AI capabilities from AWS, we significantly accelerate validation cycles and ensure consistent quality across global programmes. This solution enables us to support our customers faster and more efficiently, strengthening the foundation for the next generation of software-defined vehicles.”
IMPROVING EFFICIENCY AND CONSISTENCY
Developed with the expertise of the AWS Generative AI Innovation Centre and using Amazon Nova foundation models, Amazon Bedrock Knowledge Bases and the Strands Agents framework, the STG agent is billed as a ‘pioneering solution’ that helps Marelli to improve efficiency and boost consistency in how product features are validated against customer requirements. The tool is designed to reduce validation time and achieve stronger alignment between system requirements and validated product behaviours.
According to Marelli, these features will help vehicle manufacturers to accelerate product development for software-defined vehicles, while also delivering new functionalities with greater reliability. Designed for seamless integration with requirement management tools, the tool supports compatibility with existing automotive engineering workflows.
HOW IT WORKS
Within Marelli’s development process, customer requirements are first translated by research & development engineers into system requirements. This is a human-driven step that defines what the product is intended to do. The STG agent then analyses and identifies the expected behaviours implied by each system requirement, before automatically generating corresponding clear, structured, traceable system test cases that support Marelli’s engineers in validating that each feature behaves exactly as intended.
“Marelli’s approach to automating system validation demonstrates the transformative potential of generative AI in automotive engineering,” adds Giulia Gasparini, country leader of AWS Italia. “By leveraging Amazon Nova foundation models and Amazon Bedrock, companies are setting new standards for how software-defined vehicles are developed and validated. This solution shows how advanced AI can accelerate innovation while maintaining the rigorous quality and safety requirements that define the automotive industry.”
SOFTWARE-DEFINED STATE-OF-PLAY
According to IoT Analytics’ Software-defined Vehicles Adoption Report 2026, 45% of automotive OEMs and suppliers currently rank the transition to software-defined vehicles as their top strategic priority. The survey of over 80 automotive OEMs and suppliers found that for 45% of respondents, the transition to SDVs ranks higher than the development of both advanced driver-assistance systems (25%) and electric vehicles (14%).
According to the report, four key dimensions of software-defined vehicles are being adopted by automotive OEMs. The first is vehicle architecture, which refers to the hardware and system design that determines how computing, networking, and vehicle functions are structured. The report states the most significant architectural change in the automotive industry is the move away from complex distributed or domain-based electrical/electronic (E/E) systems, where electronics are either distributed across many independent electronic control units (ECUs) or are organise-d by function, toward two streamlined architectures: centralised zonal architecture and fully/advanced zonal architecture.
The second is vehicle-to-cloud integration; the connectivity layer that links the vehicle with cloud services to enable data exchange, updates, and remote interactions. The top role for cloud integration, as identified by automotive OEMs and suppliers in the research, is over-the-air updates (73% OEMs; 71% suppliers), where end-to-end platforms orchestrate the packaging, signing, distribution, and lifecycle management of software and firmware updates for vehicles. Other key roles include real-time data processing and analytics (68% OEMs; 73% suppliers) and collaborative design and development (59% OEMs; 45% suppliers).
The third dimension is software-driven engineering, revolving around the approach to building vehicles where software-led development, validation, and tooling shape the product lifecycle. The research found that companies that prioritise SDVs also prioritise a software-first engineering approach due to the SDV’s software-centric nature. With SDVs, the vehicle development lifecycle is becoming increasingly software-centric, with ECUs, middleware, and operating system layers designed, tested, and updated using Agile methodologies and DevOps practices. This enables faster iteration, continuous integration, and decoupling software updates from hardware production timelines.
The final aspect identified by the report is vehicle software operations and lifecycle management, which involves the ongoing processes that manage, monitor, update, and sustain vehicle software once deployed. Vehicle software operations represent a completely new operational capability for automakers. These operations are the ability to perform runtime software updates and dynamic feature deployments, all enabled by a flexible service-oriented architecture (SOA), which allows different software components to communicate and be updated independently.
