Gen AI descends into disillusionment

The excitement over generative AI in the enterprise has passed its peak, and better use cases and more accurate results may be needed to renew the enthusiasm, experts say, as the technology slides deeper into the Gartner Hype Cycle’s dreaded trough of disillusionment.

Although gen AI still has major potential for the enterprise, some expectations have diminished as organizations have run into problems with its robustness and reliability, says Birgi Tamersoy, a senior director analyst at Gartner. The hype over gen AI downplayed much of the work needed to reap its benefits, he adds.

“Instead of the news that we hear about gen AI, that we throw in data and then magically, everything happens, may not be the reality,” he says. “Proper due diligence needs to be put in place in terms of performance, robustness, and reliability.”

Hallucinations and inconsistent results from gen AI are among the major problems leading to lowered expectations, Tamersoy says. As a result, some IT leaders have struggled to find enough use cases that can tolerate the inaccuracies. At the same time, failure rates for AI pilot projects have been huge.

“AI systems inevitably make mistakes,” Tamersoy notes. “When you are building a solution, what you put around it to increase that robustness and reliability makes a huge difference in terms of success. What it will take is more in-depth evaluation of these technologies.”

Cost and energy needs have also become major considerations for enterprises as gen AI tackles more complex problems, Tamersoy adds. In some cases, energy costs can run into the millions of dollars, and organizations will sometimes need to determine whether the cost is worth the benefit, he says.

Better results needed

Other AI experts point to a lack of reliability as a big reason for deflating expectations. Gen AI too often gives users a lack of certainty, says Dmitry Mishunin, CEO of Doitong, an AI-based video platform.

“Generative AI is now a mystery box game,” he says. “You can get a masterpiece, and you can get something unusable. And even if you do get a masterpiece, nothing can protect you from getting unusable content the next time and the time after that.”

Mishunin also sees use-based pricing as an impediment. While gen AI failures would still be happen if it were free to use, certain payment models can accelerate the end of experimentation. “As soon as these services start charging us for the result rather than attempts, the industry will soar,” he says.

Gartner estimates that gen AI will take two to five years to clear the trough of disillusionment and move up the slope of enlightenment to the plateau of productivity stage in the Hype Cycle. The technology reached its peak of inflated expectations last year, Gartner contends — just as CIOs were pivoting from playtime to practicality.

Distrust in agents

Another heavily hyped AI technology, AI agents, is now at the peak of inflated expectations on Gartner’s Hype Cycle; like gen AI, it too will soon be destined for disillusionment. A lack of trust in autonomous agents will eventually drive down its overblown excitement, echoing concerns about the lack of trust in gen AI results, Gartner’s Tamersoy says.

“You cannot automate something that you don’t trust, and many of these AI agents are LLM-based right now, which means that their brains are generative AI models, and there is an uncertainty and reliability concern there as well,” he says. “If you want to automate something completely, you have to trust it very much.”

Part of the mistrust in gen AI is driven by reservations about agents, says Mike Sinoway, CEO of AI-based enterprise search provider Lucidworks. The company’s 2025 State of Generative AI report found that only 6% of e-commerce firms had partially or fully deployed one agentic AI solution, with two-thirds lacking the infrastructure to make AI agents effective.

“Faith in generative AI has slipped because expectations were misaligned around AI agents — the most promising application of gen AI,” he says. “Leaders rushed to deploy agentic solutions, trying to run before they could walk.

But as AI orchestration and agent coordination technologies develop, they will finally unlock gen AI’s potential, Sinoway claims. “The next breakthroughs won’t come from individual agents working alone,” he says. “They’ll come from orchestration systems that route tasks to the most cost-effective models.”

While waiting for new breakthroughs, Tamersoy recommends that CIOs evaluate and test gen AI tools more thoroughly moving forward to determine the right fit for their organizations. In addition, some organizations are now using multiple AI technologies together to mitigate for the inherent weaknesses in each tool used separately.

One AI technology on Gartner’s innovation trigger step, the first phase of its Hype Cycle for AI, is composite AI, which combines multiple AI techniques, such as computer vision, machine learning, and agents, to solve business problems. Tamersoy sees composite AI as one of the most promising AI technologies on the rise because of its potential to fix reliability and other problems with standalone AI tools.

“Composite AI accepts the fact that each of these techniques have their own limitations and strengths, and to build a successful solution, you may need to combine multiple techniques so that they address each other’s limitations as much as possible,” he says.

In the meantime, more model evaluation is needed, says Richard Sonnenblick, chief data scientist at strategic portfolio management firm Planview.

“We overestimated AI’s potential in the near term because we didn’t have a rubric for model evaluation and the novelty of a conversational computer charmed, or snowed, us,” he says. “But we have definitely not overestimated the medium- and long-term implications of LLMs.”

Potential still there

With better evaluation frameworks in place, steadily improving reasoning models, and well-curated data, gen AI will deliver huge efficiency gains, Sonnenblick predicts.

“The lack of faith is largely due to the dissonance between our glorious initial experiences with ChatGPT and the reality of frequent hallucinations and difficulty steering these models for business purposes,” he adds. “That said, taking risks and being fast to fail should be celebrated. Even if only one in 100 generative AI projects generate value, over time that value can justify the overall investment.”

Gartner’s Tamersoy agrees that the potential for gen AI and related technologies is still huge. “There are a lot of uses for these technologies. They can bring a lot of value, but proper due diligence needs to be put in place by organizations in terms of ensuring their solutions are performing at a high level that brings business value,” he says.