AI in Automotive Engineering: Why Smart Data Management Is More Important Than Development Speed Alone

Ahead of the International VDI Congress ELIV, the premier industry gathering for automotive electronics and software, we speak with Dr. Patrick Bartsch, Principal Technology Evangelist at AWS. At this year’s ELIV, he will moderate the Automotive Trend Session on the topic of artificial intelligence. In this interview, he explains why AI is much more than just code generation, how the role of engineers is fundamentally changing, and where companies can expect the greatest return on investment.
 

Dr. Bartsch, you will be moderating the Automotive Trend Session on automotive AI at ELIV, as well as a panel discussion on the same topic. What can participants expect?

Dr. Patrick Bartsch: We are deliberately choosing not to focus – at least not exclusively – on large language models. LLMs are undoubtedly a central and fundamental component of current technology. Thanks to what’s been called the “ChatGPT moment,” artificial intelligence has, for the first time, become accessible and popular to the general public through a simple chat window.

However, the field of AI is far broader: We’ll explore applications and agentic AI solutions based on LLMs. The key question is which specific tasks agents can autonomously handle during the development process and at which points human involvement is absolutely essential. The focus is always on concrete optimizations and a genuine return on investment for companies.
 

How exactly will AI continue to change the development process?

Dr. Patrick Bartsch: Current applications often address very practical challenges: How can massive amounts of data be searched and analyzed more efficiently, and how can extensive data lists be curated more efficiently? These processed datasets are fed into model training and subsequently find their way into the vehicle. This manifests itself, for example, in voice assistants for infotainment systems or in algorithms for autonomous driving systems.
 

This inevitably raises the question of the role of humans. Is artificial intelligence changing the job description of engineers so drastically that traditional positions will become obsolete?

Dr. Patrick Bartsch: The engineering discipline itself remains the same; only the tools change. To use a technical analogy: You can drill a hole with a hand drill or use a heavy-duty hammer drill instead. The hammer drill works faster and with more power. However, if you make a mistake, the damage caused is significantly greater than when working with a hand drill.

AI behaves in much the same way: If an agent generates 10,000 lines of code when 50 would have sufficed, errors can occur. That is precisely why we need seniority. Experienced professionals break down complex challenges into subproblems, review the architecture, and assess whether the code is secure and maintainable. Our CTO Werner Vogels puts it aptly: “There is no compression algorithm for experience.” Engineers will continue to build experience and expertise in the future by doing, experimenting, and learning.
 

A key issue in the industry is the pace of development, particularly when compared to new Asian market entrants. How do you assess the tension between the pace of development and quality?

Dr. Patrick Bartsch: My colleague Uwem Ukpong recently summed it up very succinctly: “Speed without quality is reckless. Quality without speed is irrelevant.” Focusing solely on speed will backfire in the medium term, but so will simply waiting it out.

New market entrants have the strategic advantage of operating on a greenfield basis. They aren’t burdened by legacy issues, such as having to support internal combustion engine vehicles, hybrids, and electric cars simultaneously. Furthermore, pure-play electric vehicle manufacturers are not subject to regulatory constraints such as traditional on-board diagnostics for powertrains. Established OEMs operate in a completely different context and face additional challenges, such as the ongoing development and maintenance of multiple platforms simultaneously.

The automobile is likely the most technically complex consumer product on the market and poses a high risk potential, which is why safety is the top priority. The pace set by the newcomers is impressive, but established market players possess a wealth of experience that will pay off in the long run. The key is to find the right balance between speed and perfection.
 

Can AI help established manufacturers close – or at least narrow – the development gap?

Dr. Patrick Bartsch: That is the concrete hope. Those who do not use AI will inevitably fall behind. However, a speed boost in isolation is of absolutely no use if the overarching process has not been adapted. It doesn’t help to complete a task in three weeks if the subsequent processes force a six-month wait.

Corporate culture is just as crucial. AI must not be perceived solely as a threat. When employees see that they can complete repetitive, unpopular tasks more quickly with AI support, they can channel their energy into tasks that offer real added value. AI handles tasks for which we simply don’t have the capacity at the moment. 

In customer projects, for example, we use AI to process huge volumes of test results from the cloud, perform automated image comparisons, or analyze audio clips for infotainment systems. Videos for autonomous driving functions can also be systematically scanned for algorithm errors.
 

You mention testing. The gap between simulation and reality is considered a major risk. How can the “sim-to-real” gap be closed?

Dr. Patrick Bartsch: The automotive industry will never be able to do without test vehicles or hardware-in-the-loop entirely. However, much of the work can be shifted to the cloud. To this end, it is essential to evaluate very precisely which scenarios should be simulated. Reproducing exact CAN timing exclusively in software is extremely complex, and the return on investment remains modest. However, when we move to the level of the operating system or the hypervisor of a high-performance computer, virtual simulation works exceptionally well. 

Integration issues are the primary cause of massive delays in vehicle programs. Although the number of components has decreased, the complexity has not disappeared; rather, it is now concentrated within individual devices, where processors and microcontrollers must communicate seamlessly.

The “sim-to-real” gap also varies greatly by domain. Translations in infotainment or frame rates in the instrument cluster can be validated very well in advance. However, when it comes to controlling air suspension in the millisecond range, the complex physics are decisive, and the gap becomes clearly noticeable.
 

Finally, what areas currently offer the greatest potential for automotive engineers when it comes to the use of AI?

Dr. Patrick Bartsch: A clear 80/20 rule applies here. The industry focuses heavily on pure software coding, yet in many domains that accounts for just 20 percent of the work. The remaining 80 percent of activities – such as requirements engineering, architecture design, or complex reporting – are not yet sufficiently addressed. These are highly repetitive tasks that currently have a negligible degree of automation. 

In other words: If coding output becomes four times faster, we’re only optimizing 20 percent of the process. However, if we speed up and simplify documentation, requirements engineering, and reporting, we can unlock significant efficiency potential on the 80-percent side. That’s precisely where a high return on investment lies – one that has yet to receive sufficient attention.