Capgemini: 68% of carmakers run agentic AI in product development — only 5% can govern it
The car industry has gone all-in on AI agents — and has almost nothing to show in governance. A new Capgemini Research Institute report, based on a survey of 800 automotive executives across 12 countries, finds that 68% of automotive organizations already use or pilot agentic AI in product development — while only 5% say their AI governance is fully mature.
- Capgemini Research Institute report 'Rewiring automotive product development: how organizations can pull ahead in the AI race', released in Paris on October 12, 2026, based on a survey of 800 automotive executives across 12 countries.
- 68% of automotive organizations already use (33%) or pilot (35%) agentic AI in product development — while only 5% say their AI governance is fully mature.
- Adoption in non-core engineering activities is expected to leap from 26% today to 68% within the next two to three years.
- 83% of executives believe agentic AI will let engineers spend more time on core engineering work rather than administrative and coordination tasks.
- The trust cliff: 59% trust for upstream research and concept activities collapses to 40% in development work, where AI-generated results carry direct implications for engineering outcomes.
- Ecosystem accountability gap: only 10% have established shared lifecycle accountability across their ecosystem, and just 30% report co-development models built around shared outcome metrics.
The report, ‘Rewiring automotive product development: how organizations can pull ahead in the AI race’, was released in Paris on October 12, 2026. It describes agentic AI as the next frontier in how cars get designed, engineered, and built: 33% of organizations already use the technology in product development and another 35% are piloting it. Adoption in non-core engineering activities is expected to leap from 26% today to 68% within the next two to three years.

68% in, 5% governed
The mismatch runs through the whole report. While 83% of executives believe agentic AI will let engineers spend more time on core engineering work rather than administrative and coordination tasks, governance is nearly nonexistent: just 5% of organizations report fully mature AI governance, pointing to unresolved challenges around data quality, traceability, accountability, and ecosystem collaboration.
The accountability gap extends to how carmakers work with suppliers. Only 10% of organizations have established shared lifecycle accountability across their ecosystem, and just 30% report co-development models built around shared outcome metrics — even as software-defined vehicles push development across OEMs, suppliers, and technology partners.
The trust cliff at safety-critical work
Executives are comfortable with agents in the early stages — trust reaches 59% for upstream research and concept activities — but that confidence collapses to 40% in development work, where AI-generated results carry direct implications for engineering outcomes, product quality, and safety. More than seven in ten executives believe agentic AI can improve engineering decision-making, yet confidence drops significantly as AI moves closer to safety-critical activities.

The winners aren't just adopting faster — they're better organized
A select group of “high-value realizers” is pulling away: they report 17% faster development cycles, 19% gains in engineering productivity, and a 19% reduction in time-to-market — up to three times the gains of lower-performing peers. Their advantage, the report says, comes less from adopting AI itself and more from stronger data foundations, clearer business cases, and more disciplined governance and operating models.
“The organizations generating the most value are not simply adopting AI faster, they are redesigning workflows, strengthening ecosystem collaboration, building trusted data foundations and establishing clear accountability for AI-enabled decisions,” said Laurence Noël, Global Head of Automotive at Capgemini. “The leaders of tomorrow will be those that successfully embed AI into the engineering processes at scale.”
The vehicle feedback loop that doesn't close
The report also flags a problem beyond AI adoption: carmakers struggle to turn real-world vehicle data into continuous improvement. Nearly two-thirds (64%) say learnings from vehicles in use mainly inform future vehicle programs rather than improving the cars already on the road. Over-the-air updates remain concentrated on cybersecurity and navigation, with far fewer organizations deploying updates that modify driving functions or assisted-driving capabilities.
The barrier is organizational, not technical
Six in ten executives say delays are driven primarily by process inefficiencies rather than technology limitations — and 45% have yet to define a strategy and roadmap for AI in product development at all.
Maturity also varies sharply by market. Among OEMs in the US and Europe, 53% and 46% are already implementing AI initiatives, with 18% and 16% running a comprehensive cross-functional AI strategy. In China, four in ten respondents are in the implementation phase, but only 7% report comparable strategic maturity.
Methodology: the Capgemini Research Institute surveyed 800 senior executives — 400 automotive OEMs, 160 tier-1 suppliers, and 240 technology providers — across North America, Europe, and Asia-Pacific, and conducted 24 in-depth industry interviews. The survey ran in July 2026.
Sources: Capgemini press release via GlobeNewswire, Oct 12, 2026; Capgemini Research Institute.
Frequently asked questions
What did the Capgemini survey actually find?
Based on 800 automotive executives across 12 countries: agentic AI is moving from pilots into the engineering core — 33% already use it in product development, another 35% are piloting — but oversight is not keeping pace, with only 5% reporting fully mature AI governance. Governance gaps center on data quality, traceability, accountability, and ecosystem collaboration.
What is the 'trust cliff'?
Executives are comfortable with agents in early stages — trust reaches 59% for upstream research and concept activities — but confidence collapses to 40% in development work, where AI-generated results carry direct implications for engineering outcomes and safety-critical decisions.
How fast is adoption expected to grow?
Adoption in non-core engineering activities is expected to leap from 26% today to 68% within the next two to three years, while 83% of executives believe agentic AI will let engineers spend more time on core engineering rather than administrative and coordination tasks.