The automotive industry is facing a far-reaching turning point, as artificial intelligence is becoming increasingly integrated into the core processes of vehicle development. At this year’s VDI Congress ELIV, Dipl.-Ing. Steffen Krause, Head of Software Defined Vehicle at Capgemini Invent, therefore aims to usher in a shift in perspective. In his in-session keynote with Bora Ger, also of Capgemini Invent, he makes it clear: AI is not only transforming automotive development – it is upending the entire established playbook. Ahead of the event, he answers our questions.
Mr. Krause, how radically is AI currently changing vehicle development?
Steffen Krause: Across the industry, we’re seeing a massive push to use AI to accelerate established processes. The approach here is for companies to apply AI “patches,” so to speak, to individual process steps. In specific cases, this works exceptionally well. However, when you ask senior management about the substantial value that AI adds to the company as a whole, the response is often sobering.
The reason for this lies in a massive logical disconnect. If an AI model generates millions of lines of code overnight, but the subsequent review process continues to rely on traditional manual human inspection, the system will inevitably become overwhelmed. No developer in the world can seriously validate this enormous volume the next day. The answer, however, is not even more manual review, but a paradigm shift: We must consistently pair AI-generated code with automated verification – AI-generated artifacts are checked against formal specifications instead of being manually reviewed line by line. Furthermore, we must shape the collaboration between humans and AI in this process, which we refer to as “Human-AI Chemistry.”
This shows that if you focus solely on drastically accelerating a single aspect without considering the overall process, you won’t create real value – you may instead create a massive bottleneck. This is precisely the critical juncture we find ourselves at today when it comes to integrating artificial intelligence. We must move away from isolated pilot projects and toward end-to-end, integrated AI engineering pipelines.
To gain a concrete understanding of this rapid change, you have collected extensive data. At ELIV, you’ll provide insights into a reality check for the year 2026. Where do manufacturers and suppliers currently stand?
Steffen Krause: We regularly conduct comprehensive surveys in the run-up to ELIV. For the current study – which we are conducting jointly with ASIMI and in cooperation with the Technical University of Munich – we surveyed approximately 16 senior executives and technical decision-makers from across the entire development ecosystem, ranging from OEMs to software Tier 1 suppliers, system integrators, and hyperscalers, all the way to tool providers and research institutions, regarding the maturity level of their AI integration. The data we will present to the public around the time of the conference paints a very clear picture. Many companies are still in the phase of purely ad hoc automation.
We are by no means pursuing the utopian vision of AI independently designing entire vehicles in the future. Rather, the core task is to make the interaction of digital tools measurable and controllable. The study provides us with a solid foundation to show the industry where urgent action is needed and just how wide the gap between technical feasibility and process reality currently remains. With this, we aim to provide clear impetus for a new target operating model within companies.
As is well known, the automotive industry operates in an extremely heavily regulated environment. The development process always culminates in regulatory type approval. Is this regulatory rigor compatible with the exponential pace of AI models?
Steffen Krause: This is one of the most exciting challenges that the new playbook for AI-based vehicle development will bring. Complex vehicles are being created for which the manufacturer will ultimately bear full legal responsibility. We all appreciate the fact that a car brakes with absolute reliability when we press the pedal. This uncompromising safety is the top priority. Consequently, regulatory authorities must adapt to the new pace of development, while manufacturers must invest heavily in reliable AI-supported approval mechanisms. Crucial to this is the necessary shift in perspective regarding the use of these tools: trust is not built by waiting for a supposedly error-free or “deterministic” AI system.
Another, often underestimated aspect is geopolitics. Anyone who aligns their business-critical development processes exclusively with a single frontier model exposes themselves to avoidable dependencies. Just recently, a leading U.S. frontier model was placed under U.S. export controls on short notice and was unavailable for several weeks before access was restored – a vivid example of how quickly regulatory frameworks can change. Conversely, while Chinese open-source models such as DeepSeek do not pose commercial licensing hurdles – their current flagship models are licensed under permissive licenses – they do raise unresolved questions regarding data protection and GDPR compliance, data residency, and the enforceability of the EU AI Act. If an automaker builds its core process on such models without safeguards, regulatory changes could bring the entire development process to an immediate standstill. What is crucial, therefore, is not isolation but flexibility: rigorous risk management, a neutral integration layer that allows models to be swapped out as needed, and the deliberate freedom to choose from a broad range of open-source and proprietary models, both European and international. In this sense, sovereignty means business continuity and risk management, not isolation.
Under the term “Human-AI Chemistry,” you’re calling for a consistently data-driven approach. But let’s turn to the human component. Will AI eventually make our engineers obsolete? Junior positions, in particular, are considered to be at acute risk.
Steffen Krause: This fear of automation is nothing new; we’ve seen it in every major industrial revolution. However, the idea that we could do without junior developers entirely in the future and work exclusively with senior experts and language models is blatantly wrong. Such a rigid setup would fail spectacularly and wouldn’t last two years.
What is true, however, is that specific job roles are changing drastically. AI is taking over the purely boilerplate code, the tedious creation of routine presentations, and the drafting of administrative tickets. At the same time, we need the human touch more than ever. Translating the strategic guidelines of an experienced senior into a highly precise dialogue with the AI model and shaping that into a truly value-adding result remains an extremely challenging task for junior staff. It is precisely through these daily tasks that they gradually grow into their future senior roles.
At the same time, the experienced engineer, who has a detailed command of complex control technology for brake control units, finally regains the freedom needed to focus on his core engineering work, rather than spending valuable time on mandatory administrative tasks. It is precisely this interplay that we at Capgemini bring together in our Resonance Framework under the dimension ADOPT – the “Human-AI Chemistry”: Trust, clearly defined roles, and well-designed interactions are the prerequisites for “people who use AI” to become “people who work with AI” and for hybrid teams to truly function effectively.
This technological realignment therefore also requires the appropriate financial budgets. Do all decision-makers already understand how to measure and evaluate the return on investment for AI projects in a sustainable manner?
Steffen Krause: In fact, the way work performance is evaluated is also undergoing a fundamental shift right now. Nvidia CEO Jensen Huang recently proposed making employees’ token consumption a productivity metric – even going so far as to include token budgets as part of compensation. For the automotive industry, I believe this is the wrong metric. The hardware supplier may have a strong interest in high token consumption, but in the industry, the only thing that counts is the created, validated outcome, not the volume. For years, business metrics were strictly based on hours worked. But that no longer works in an AI-driven world. In the future, the value of a complex development will be measured by the created, validated outcome, not by the hours invested.
Therefore, investments must be consistently aligned with how much faster measurable value is generated. This forces companies to make profound adjustments to their procurement strategies, FinOps processes, and overall sourcing models. Those who manage purely based on token consumption will inevitably lose sight of true value creation. It is important to note that the actual return does not come from the tool alone. Value arises from the combination of technology and (human) processes, not from the tool itself.
Whenever the pace of development in the automotive industry is discussed, the term “China Speed” inevitably comes up sooner rather than later. What is the reason for this rapid pace of development? Is it simply because Asian competitors take a less perfectionist approach to development?
Steffen Krause: First, a clarification: The so-called “China Speed” was not a formal part of our survey – I’m including it here as a personal assessment and for context, not as a study finding. The widespread myth that “China Speed” results purely from a relaxed, iterative approach falls far short of the mark. Rather, the decisive competitive advantage of the new players is the famous “blank slate.” These companies are not burdened by limiting legacy software or historically evolved, highly complex supplier networks that force them into lengthy coordination processes.
In contrast, there is the undisputed strength of German premium manufacturers. As is well known, no customer has ever complained about the outstanding mechanical quality of a Porsche. The key lies in maintaining these classic strengths and continuing to evolve based on the brand’s core values. The immense challenge – and at the same time a major opportunity – for Europe as a business location now lies in absolutely preserving this mechanical perfection and combining it with software agility in the future. AI offers exactly the tool needed to quickly close these existing gaps in processes.
Final question: In your opinion, what options do established premium manufacturers still have to clearly position themselves and differentiate themselves in the market?
Steffen Krause: Clearly, emotions, brand, and intuition. Imagine an artificial intelligence designing a new vehicle entirely on its own. Who would it be doing that for? Probably for another AI. Such a purely logical vehicle wouldn’t need comfortable seats – and it definitely wouldn’t need an emotional design. We build cars for real people.
The real art lies in seamlessly integrating these new technological accelerators with human developers: People remain in charge and orchestrate the process, while the way work is done evolves – moving away from sprints and release trains toward specification- and model-driven development, in which agents operate within clear guidelines. Whoever is the first to perfect this interplay across the entire value chain will shape the future market. The automotive industry has traditionally had a remarkably steep learning curve, and the new playbook provides the roadmap for the next major technological leap.
