Engineering transformation
Agentic AI exposes governance gaps in vehicle development
AI is already established in vehicle development, but many companies still lack the organisational framework needed to govern its use.
Mercedes-Benz
Agentic AI is moving rapidly into vehicle development, but organisational controls are struggling to keep pace. A Capgemini study finds that almost seven in ten automotive companies already use or pilot the technology, while governance, trust and collaboration remain major gaps.
Agentic AI is becoming an essential part of automotive product development. According to a study by the Capgemini Research Institute, 33 per cent of companies already use it in production, while another 35 per cent are running pilot projects. Almost seven in ten automotive companies are therefore deploying or testing agent-based AI systems.
Expectations are rising quickly. Some 83 per cent of executives believe agentic AI can relieve engineers of administrative work and free up more time for core development tasks. In indirect engineering functions, usage is expected to rise from 26 per cent today to 68 per cent within the next two to three years. Agentic AI systems across automotive engineering workflows are moving rapidly from trials towards wider adoption.
How are AI leaders pulling ahead?
The study points to a widening performance gap. Companies that have already integrated AI successfully into development report 17 per cent faster development cycles and 19 per cent higher productivity, alongside a 19 per cent shorter time-to-market.
The difference lies not simply in adopting AI, but in embedding it within the organisation. According to the study, higher-performing companies combine robust data structures, clearly defined use cases and suitable governance models. This reflects a wider move towards scaling AI across processes, toolchains and governance instead of treating applications as isolated experiments.
“With the transformation towards the software-defined vehicle, the automotive industry has undergone a profound change in recent years. This created the digital foundations for the next wave of innovation,” says Capgemini expert Heiko Huettel. “Agentic AI is now creating a new point of differentiation across the industry.”
Why is governance failing to keep pace?
Only five per cent of the companies surveyed have fully mature AI governance in place. At the same time, 45 per cent have yet to define a robust strategy or roadmap for using AI in product development.
Trust also falls as AI moves closer to safety- and quality-critical decisions. While 59 per cent of executives trust agentic AI in early research and concept phases, that figure drops to 40 per cent when AI-generated results could directly affect product quality or safety.
“Rapid AI adoption does not automatically lead to success,” Huettel says. “To create real value, companies need to redesign workflows, strengthen collaboration with partners, establish robust data foundations and define clear responsibilities for AI-supported decisions.” Governance is therefore becoming an organisational challenge as much as a technological one.
How much vehicle data still goes unused?
The study also identifies weaknesses in the use of vehicle data. Connected vehicles generate large volumes of operational information, but many manufacturers still apply those insights mainly to future vehicle generations. Some 64 per cent of respondents say data from real-world operation is only rarely fed directly into the optimisation of vehicles already on the road.
That leaves a gap between data collection and continuous product improvement. Real-world driving data can feed development and validation, but this feedback loop is still far from routine across the industry.
Over-the-air updates remain focused mainly on cybersecurity and navigation functions. Driving functions and driver-assistance systems are improved through software updates much less frequently.
What is blocking AI at scale?
For many companies, the main obstacles to scaling AI are now organisational rather than technical. Six in ten executives identify inefficient processes as the leading cause of delays in AI projects. Collaboration across the value chain also remains fragmented.
Only 30 per cent of companies have collaborative development models with shared performance indicators, while just ten per cent have established joint responsibility across the full product lifecycle. The findings suggest that a different kind of collaboration across the automotive ecosystem will be needed if agentic AI is to move from isolated use cases into industrial-scale engineering.
Capgemini’s findings therefore suggest that the next competitive divide may be shaped less by access to AI than by governance, data integration and organisational design.
