Gartner: GPT-5 is here, but the infrastructure to support true agentic AI isn’t (yet)
1 min read
Summary
Excitement and investment in AI have reached new highs with the incremental improvements witnessed in language models such as Open AI’s GPT-5, but the technology is closer to the “Peak of Inflated Expectations” phase of Gartner’s Hype Cycle ahead of the “Trough of Disillusionment” than many analysts are prepared to admit, sending a strong signal to the market that the AI agents that everyone has been waiting for still remain some way off in the future.
Progress requires a complete reinvention of model design and architecture, and a revolution rather than evolution, with today’s AI still not close to being able to pass what Gartner calls the “plasticity test”, or the ability to learn rapidly in domains with limited data.
A deficit of adequate infrastructure is also holding back progress, with AI agents lacking access to a comprehensive set of suitable enterprise tools and the ability to interface with data stores and software as a service apps, as well as the necessary identity and access management systems to control agent behaviour and access and oversee the types of data that they can access.