"Generative AI (artificial intelligence) is largely hindered by the so-called '80% accuracy wall,' where most models achieve only around 80% accuracy and thus fail to impact corporate profitability. This is because they cannot be deployed on-site without resolving the remaining 20%. Makinax has successfully operated in environments where even a 1% margin of error is unacceptable, and continues to do so today."
Yoon Sung-ho, CEO of MakinaRocks, made these remarks at 'Attention 2026' held on the 3rd at Western Seoul Parana in Gangnam-gu, Seoul. Attention is an annual industry AI conference organized by MakinaRocks since 2024, marking its third edition this year. Following last year's attendance of approximately 500 participants, this year's event drew around 700 attendees from the manufacturing, energy, semiconductor, defense, and public sectors.
During the event's opening session, he cited a survey by the American Economic Research Institute indicating that 90% of companies have yet to experience productivity improvements through AI. He also referenced McDonald's case, where a three-year trial of AI in drive-thru operations failed due to frequent errors. This illustrates how few real-world applications of AI exist in manufacturing settings due to accuracy concerns.
Yoon (CEO) said, "MakinaRocks focused on creating AI that operates on the ground, surpassing this '80% barrier,'" adding, "The answer lies in verifying and analyzing field data and understanding business processes." Just as the saying "Umunhyeondap" (Our answers lie in the field) suggests, improving AI accuracy also begins with identifying the actual sites where AI will be applied.
As a result, over 6,000 AI models are now actively operating in industrial settings. Key examples include: △reducing annual review time for design document changes by more than 1,000 hours; △applying AI to over 1,400 robots across six automotive production plants; and △participating in the first U.S.-South Korea joint training (UFS) that piloted AI integration into military battlefield networks.
The core competitive advantage lies in Makinax's robust FDE (Folded Deep Field Engineer) organization. AI engineers within this team have invested 70,000 hours visiting over 80 client sites across 32 cities in five countries to provide close on-site support. Additionally, the company is collaborating with electric actuator firm EnerTock to develop fully autonomous factories.
Yoon (CEO) identified 'corporate AI sovereignty' as an essential condition for building AI competitiveness. Corporate AI sovereignty is defined by three pillars: ownership, independence, and control. 'Ownership' ensures that proprietary corporate data, business processes, expert judgment criteria, and AI learning outcomes remain internal assets without leaking outside the company. 'Independence' means maintaining the flexibility to replace or combine models and computing environments as needed, without dependency on specific foundation models, GPUs, or cloud infrastructure. 'Control' involves recording and managing every step of the AI process: what data was used for decisions, what actions were taken, and how much cost and resources were consumed. He explained, "Our operational sites will one day extend beyond Earth. Our goal is to build factories on Mars through automation."
Meanwhile, Attention 2026 featured presentations from representatives of various companies including Daim Research, Korea & Company Group, HD Hyundai, Samsung Electronics, KIST-Eji Institute, HD Hyundai Heavy Industries, Samsung SDS, and Dwelve Labs, who shared their AI implementation cases.
in STOCK
