Automakers are increasingly relying on software to introduce new features and improve vehicle functionality. According to McKinsey, the global automotive software and electronics market could reach $519 billion by 2035. Yet growing complexity creates new challenges, particularly around compatibility across platforms. In March 2026, Google announced Android Automotive OS for Software Defined Vehicles (AAOS SDV), an initiative aimed at reducing fragmentation through more standardized software infrastructure.
However, standardization alone cannot guarantee reliability. Engineers still need to validate complex integrations, identify defects early, and ensure that updates do not disrupt existing functionality. AI can help automate these processes, from generating tests to investigating failures. However, AI-generated solutions can also introduce errors or overlook critical issues, making human oversight essential.
Inna Olenina brings more than 25 years of experience to these challenges. She is now a Senior Software Engineer at SiriusXM, one of North America's leading audio entertainment companies. In this interview, Olenina explains how engineers can address the growing challenges of software integration, testing, and reliability.
Inna Olenina possesses many years of experience in the development of mobile and embedded software. She has also made major contributions to projects for major manufacturers such as Motorola, Nokia, Samsung and Sony Ericsson. She worked on user interface and application development as well as on operating system technology, including MeeGo for Nokia devices.
She went into medical technology and developed software for Stereotaxis, a company that produces robotic systems to treat heart conditions. One of her key contributions was an application that controlled multiple video sources and connected systems.
Another major milestone was her work on a large-scale retail auditing solution for Nielsen. The applications she helped develop were used by tens of thousands of auditors across more than 70 countries to collect retail data. Beyond development, Olenina provided post-deployment support. She worked directly with international users to resolve technical issues and coordinated efforts across development teams in different countries.
"In medical equipment, I had to understand how software interacts with hardware, as well as the strict requirements for reliability and accurate data processing. In retail, the focus was on large volumes of data, business logic, and reporting," Olenina explains. "The underlying engineering principles remain largely the same. What changes is the domain knowledge you need to apply them effectively."
This ability to adapt established engineering practices to unfamiliar technologies has proved particularly valuable at SiriusXM, where Olenina has taken on new challenges involving satellite data, automotive software, and artificial intelligence.
One of Olenina’s biggest projects at SiriusXM was a web application for visualizing and analyzing data from the company’s satellite information services. They provide weather forecasts, radar images and other meteorological information to pilots and vessel operators. This application plays an important role in maintaining quality of service.
It lets engineers review reported problems, re-create weather conditions for specific times and locations, and validate displayed data using interactive maps and historical data analysis.
Olenina led the development of key visualization features and solved some of the most difficult performance problems in the project. She built interactive mapping components, expanded support for different data types, and streamlined the handling of frequently updated information to keep the interface responsive.
"When large volumes of data arrive, they need to be decoded, retrieved from the database, and displayed on the map without delays," Olenina explains. "I optimized data retrieval and filtering, reduced unnecessary processing and rendering, and worked on asynchronous operations and caching."
Olenina also incorporated AI-assisted tools into her workflow to identify potential defects, explore alternative solutions, and improve code quality. Rather than taking AI-generated recommendations at face value, she carefully validated them through testing. Together with her performance improvements, this approach helped make the application faster and more reliable.
Her next project took Olenina from satellite data to automotive software, where she tackled a new challenge: making testing more reliable across different vehicle platforms.
Modern cars rely on a complicated mix of software pieces, services and hardware interfaces. Even when manufacturers are using Android Automotive OS, differences in configurations can cause compatibility issues. An app that works fine in an emulator may fail on a real head unit, the computer inside a car's dashboard. Security restrictions on these devices may also reduce the usage of conventional testing tools, making validation more difficult and time consuming.
To address these issues, Olenina has designed and implemented a unique and extensible automated testing framework for Android Automotive OS. As the project's architect and principal developer, she was responsible for its core technical decisions and implementation.
The framework supports API, integration and user interface testing on head units and emulators. This allows engineers to run automated checks on real hardware, but still within the security constraints of each platform. It also uses controlled test data with pre-determined expected results, ensuring that the testing is consistent and reliable, and does not rely on live content, which can be unpredictable.
Another major innovation is the possibility to validate interactions through a platform abstraction layer (PAL), which links applications with vehicle-specific systems. Historically, such integrations have involved a lot of manual verification. Olenina’s method cuts down on repetitive work, while also helping engineers to spot compatibility issues earlier.
Thanks to its modular architecture, the framework can accommodate new testing capabilities without major changes. One of its latest advances is automated UI testing on head units. "The main challenge was not simply to automate individual tests, but to create an architecture that could work across different automotive environments," Inna Olenina explains. "I wanted to build a framework that could test these integrations on actual equipment, produce predictable results, and be extended without redesigning the entire infrastructure."
Another distinctive feature of Olenina's framework is its use of artificial intelligence. AI tools analyze technical specifications, source code, and existing tests to identify gaps in coverage and suggest new test scenarios. They also help investigate failures by examining results, application logs, and documentation to pinpoint possible root causes. However, Olenina draws a clear line between AI assistance and human oversight. “Engineers still verify AI recommendations. AI does not make decisions independently,” she emphasizes.
The framework has already streamlined repetitive testing, improved reproducibility, and enabled more frequent regression runs. Based on representative scenarios, automation could reduce manual effort by an estimated 65-85%, saving approximately 2.5-3.5 engineering hours per cycle, or 35-50 hours across 14 typical cycles.
As the project grew, Olenina took on the leadership of a three-engineer team responsible for developing and executing tests using her framework. She now oversees their work while continuing to shape its architecture, expand its functionality, and guide key technical decisions.
The challenges addressed by Olenina's framework reflect a broader transformation in the automotive industry. As vehicles become increasingly software-defined, manufacturers can introduce new capabilities and improve existing functions throughout a car's lifecycle. This shift, however, requires a fundamental change in how automotive software is designed, tested, and maintained.
One major development is the move toward modular architectures and standardized interfaces. Google's AAOS SDV initiative represents an important step in this direction. It introduces a service-oriented architecture, common communication mechanisms, and support for virtualized development environments such as Cuttlefish.
"The main value of Google's initiative is the opportunity to standardize part of the vehicle's software infrastructure, allowing engineers to focus more on functionality rather than the specifics of each platform," Olenina explains. "This could enable the reuse of software components, test scenarios, and validation tools. I see AAOS SDV not as a universal solution to every challenge in automotive software development, but as a significant step toward a unified software ecosystem."
Standardization is also crucial to long-term software maintenance. Over-the-air (OTA) updates allow manufacturers to improve vehicles long after they leave the factory, but they also bring new challenges in version management, compatibility, secure deployment, and rollback. As updates become more frequent, engineers must ensure that changes to individual components do not compromise the reliability of the system as a whole.
Meeting these demands requires more efficient approaches to testing. Virtualized and cloud-based environments allow specialists to identify defects earlier and run tests at scale, even before physical hardware becomes available. However, real devices remain essential for validating hardware-specific behavior, performance, and safety-critical functions. Olenina advocates a hybrid approach that combines both methods — a principle reflected in her own testing framework.
Artificial intelligence is also reshaping engineering workflows. Modern AI agents can tackle increasingly complex development tasks, from analyzing large codebases to making changes across multiple components. Yet Olenina cautions that reliability remains a major limitation, particularly in lengthy, multi-step projects or when requirements are incomplete or unclear.
"AI can propose a solution that compiles successfully and even passes tests while still containing architectural flaws, overlooking important edge cases, or breaking existing functionality," she explains. "This is especially concerning in embedded systems, where critical hardware constraints may not be fully reflected in the code or documentation." In Olenina's view, AI is most effective when applied to well-defined tasks whose results can be objectively verified. Her own testing framework puts this approach into practice, combining AI-assisted automation with rigorous human oversight.
As AI takes over more routine development tasks, Olenina expects engineers to shift their focus toward system design and evaluating AI-generated solutions. "The value of an experienced engineer will increasingly lie not in how quickly they write code, but in their ability to make sound architectural decisions," she says. "Understanding the broader system, weighing technical trade-offs, anticipating risks, and ensuring reliability will still require human expertise."
This places greater importance than ever on strong foundations for aspiring developers. Olenina tells them to learn algorithms, data structures, operating systems and software architecture. Embedded development requires a good understanding of memory management, multithreading, hardware interfaces, and real-time systems.
Her other advice for young engineers is to use AI tools, but not to be dependent on them. Instead of passively accepting the solutions generated, they should seek to understand the rationale behind each recommendation, evaluate potential limitations and alternative approaches, and develop the ability to independently troubleshoot intricate problems. Hands-on projects that require them to design, test, debug, and deliver working software remain one of the best ways to develop these skills.