The broad spectrum of Agentic AI

Our last edition highlighted a fundamental lack of data excellence as a critical limiting factor to successful AI adoption. This time around we’re exploring the broad spectrum of Agentic AI and interrogating topics such Secure AI and what that means for organisations, especially those already engaged in progressive forms of AI and multi-agent solutions.

There’s been a lot of recent media coverage around frontier models and the calls from technology and political leaders to ‘pace the frontier’. AI models today (including open-source options) are already able to deliver multiple trillions of dollars of value to organisations globally, the real need is how to add more control and govern the deployment of AI, not just the development of models themselves. What’s becoming evident is that agent architecture, model selection and notably how to deploy AI securely, govern and optimise it’s use effectively are emerging with a level of importance greater than that of the raw model itself. This is going to be a topic we return to frequently, both as an organisation and industry.

In the meantime, the landscape continues to evolve rapidly. Studies estimate that enterprise adoption of agentic AI reached 31% in early 2026, with projections suggesting as high as 48-55% by early 2027. In contrast, however, research from Gartner shows that 40% of agentic AI projects will be cancelled by 2027 due to unclear ROI and inadequate governance structures. The organisations with the competitive edge are those embedding agents into defined workflows with measurable business outcomes, not those attempting enterprise-wide deployments without a clear plan which are increasingly resulting in cancelled projects.

There’s a lot to unpack so I hope you enjoy this edition of AI Horizon and find the perspectives both interesting and helpful.