[Digital Today reporter Chi-gyu Hwang (황치규)] IT market research firm Gartner forecast that more than half of generative AI projects will exceed budgets due to poor architecture choices and a lack of operational know-how. Gartner also expects most companies seeking to build their own customised models to abandon those efforts because of cost, complexity and technical debt, The Register reported on May 28 (local time).

According to the report, Gartner assessed 30 AI technologies in its "2026 Generative AI Hype Cycle" report and said none had reached the Plateau of Productivity. The plateau is a phase in which a technology stabilises after 2 to 3 generations of evolution and delivers verified, practical effects.

Technologies climbing the Slope of Enlightenment include generative AI application apps such as coding assistants, graphics and video generation, and content summarisation. Concerns over intellectual property and inaccurate outputs remain issues, but foundation models are advancing rapidly and more than half of target markets have already adopted them.

Domain-specific generative AI models deliver better results than general-purpose models in areas such as healthcare, finance and law, and they hallucinate less. They require significant computing resources, specialised staff and ongoing maintenance. Gartner assessed the maturity of the market as "adolescence" and forecast it would take 2 to 5 years for mainstream use.

AI agent communication protocols were assessed as the least mature technology. MCP and A2A are the best known, but alternatives are emerging and all are evolving quickly, The Register reported.

Gartner identified 2 technologies as having the greatest potential. One is "disinformation security" tools that counter disinformation attacks using deepfakes and impersonation, though it forecast 5 to 10 years for maturity. The other is "world models" that abstract the physical environment so AI can perform prediction and planning tasks. They are also useful for robot guidance and for generating AI video that accurately reflects physical laws.

Gartner said that to use top-tier technology in open-source AI models, it may be necessary to consider Chinese models. It analysed that as U.S. companies selectively disclose their own models, leadership in open-model innovation has shifted to China.