"AI can help us overcome the so-called 'magic triangle,' the tradeoff between speed, cost, quality, and innovation."

Vector Korea (CEO Ji-Hwan Jang), a global vehicle software solutions company, held 'Vector TechDay Korea 2025' at the Conrad Hotel in Yeouido, Seoul on the 26th.

At the event, Vector Director Christof Ebert delivered a keynote speech titled 'GenAI from Requirements to Testing.'

“In SDV, we must prioritize both speed and functional safety,” says Christof Ebert, Managing Director of Vector. “Speed alone is not enough. “We must maximize efficiency by combining traditional values and the latest technology,” he said.

The concern of companies that it is easy to improve only two of cost, quality, and innovation, but difficult to achieve all three simultaneously is called the "magic triangle."

On the other hand, the solution is to remove unnecessary complexity.

Real customers only value half of a product's features.

The remaining 50 percent is just a waste of time and money, with unnecessary deformation points and repetitive processes.

“We need to determine real value from the requirements stage,” said Vector Managing Director Christof Ebert, dividing the AI strategy into three axes: product development, work environment, and customer value.

In particular, the role of requirements and test automation tools was emphasized in terms of 'customer value'.

“AI is a technology with a 70-year history, but it is now overly hyped,” he warned, adding, “Especially in the area of code generation, there is a constant risk of copyright infringement, security vulnerabilities, and malicious code insertion.”

He reiterated the importance of test automation, stating, “Testing is as critical a task as design and coding.”

It was suggested that a test strategy should be implemented in parallel during the design phase to ensure that requirements are verifiable.

To this end, Vector introduced the concept of 'central requirements engineering'.

AI automatically reviews requirements for comprehensibility, testability, and compliance with standards.

Vectors are the core elements of domain knowledge.Take it as a reward.

“Precise domain-specific knowledge improves the performance of language models, reduces energy consumption, and allows smaller models to suffice,” explains Christof Ebert, Director of Vector.

The AI tool is designed to act as a "co-pilot" to assist developers, allowing test engineers to take the lead in reviewing the results.

To prevent quality degradation during iteration, a human review process is essential.

Removing duplicate test cases, which account for 40-50 percent of testing costs, and automating the identification of "corner cases" such as boundary and exception conditions were also introduced as specific examples.

Corner cases can be derived using a procedure similar to SOTIF (safety) analysis.

In a case study of landfill drum detection conducted in collaboration with university researchers, synthetic data was generated to reduce testing efficiency from 40 days to approximately 100 cases per day.

Manual verification is still required, but speed has been significantly improved.

“AI-based automation must be predicated on the right methodology to increase efficiency and speed while maintaining quality,” said Vector Managing Director Christof Ebert. “We must not forget that the ultimate responsibility lies with people.”

Finally, he emphasized the importance of acquiring new AI capabilities.

Vector Director Christof Ebert urged, "Co-pilot users must be equipped with critical review skills. Uncritically accepting the generated content can lead to major accidents."