Stuck In A Rut

Arguably, the most interesting and pivotal stage in the process is the one Gartner says GenAI has now reached: stage three, the “Trough of Disillusionment.” This is where expectations are assessed against real-world outcomes, and potential should be giving way to impact.

In the case of GenAI, it’s easy to see why the hype train has picked up so much speed. But extreme success also brings risks: Its adoption has been both top-down and bottom-up, and, in many cases, has been driven by the prospect of short-term wins and FOMO rather than by tried-and-tested design and implementation processes. For enterprises where technology doctrine is driven by caution and control, this goes completely against the grain.

As a result, AI usage has become decentralized and inconsistent, with different parts of the business using different tools for similar tasks and with varying levels of oversight. Even more concerning is the gap between what leaders believe is happening and how AI is actually being used on a day-to-day basis. Organizations have a major blind spot around prompt-level interactions and the way data is being shared across AI systems and beyond.

Dangerous Assumptions

But how does this manifest itself? In many cases, leaders rely on assumptions or incomplete data about how their teams actually use AI. They are also under significant pressure to demonstrate that their use of AI is responsible, with some organizations actually stifling innovation by imposing restrictive policies that limit how and where AI can be used.

The inevitable consequence is a growing disconnect between investment in AI and the ability to measure its impact in any meaningful way. This needs to change so leaders have a much better understanding of how AI is being used at every touchpoint.

In most organizations, the evidence is already there, even if it’s probably not being captured or acted upon. Indeed, every interaction with AI generates a detailed record of behavior, including prompts, workflows, tool usage and agent activity. It is all signal that can be understood and made actionable.

And let’s be clear, this isn’t just about control. When organizations are able to access and interpret this data, they open up a much clearer view of which tools are driving productivity and which are not. They can also identify emerging use cases based on real behavior rather than assumptions, as well as areas where demand is not being met by existing tools or processes.

This also provides important context for risk, allowing organizations to understand not just where exposure exists but also how and why it occurs. Crucially, this shifts the conversation away from simply controlling AI usage to learning from it and using that insight to guide future decisions.

Final Thoughts​

These are not trivial considerations. Clearly, both generative and agentic AI will negotiate the hype cycle; of that, there is no doubt. But there will be enterprise casualties along the way, and underperformance is increasingly entering the conversation, with one widely quoted MIT study from last year concluding that “95% of generative AI pilots at companies are failing.”

Get it right, however, and there is an unprecedented win-win on the horizon of transformative AI potential that translates into reality.​​

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?