Kim Ki-dong (김기동) of AI and data specialist Encoa said differences in AI agent results come from data readiness, not the model. He said many firms saw strong results in PoC and demos but struggled in real operations because legacy data lacks clear meaning, links and context. He defined AI-ready data as data AI can understand and trust. Encoa implements this with a “context map” and proposes semantic models and two graphs to find data and generate answers.

Dinotesia said it unveiled a server configuration using four processor cards for vector search at the AI Infrastructure Summit 2026 in Santa Clara. The company said an FPGA-based test showed vector-search throughput of up to 5.77 times that of a dual-socket CPU-only server running the same software stack. In a 4,096-dimension multimodal workload, host CPU usage fell 92 percent and memory use 73 percent.