TLDR
- Enterprise demand for AI remains strong, but adoption maturity varies widely, creating “hype fatigue” among executives, according to Jim Rowan, Deloitte’s U.S. head of AI.
- Accuracy is improving and costs are falling as companies shift to multi-agent architectures and lightweight models, though governance challenges are rising.
- Deloitte warns that a rapid capability leap could force companies to transform quickly, leaving late adopters struggling to catch up.
As Wall Street sells off tech stocks over fears of an AI bubble, Deloitte’s Jim Rowan remains focused on the long game. From his vantage point working with corporate clients, he sees enterprise appetite for AI applications and compute remaining strong and showing little sign of tapering.
Rowan, the U.S. head of AI for the consulting firm, told The AI Innovator that the majority of Deloitte’s clients are in various levels of AI adoption – from those deploying their first AI chatbots while they sort out policies, procedures and data access, to those already jumping into AI agents for revenue growth.
Across industries, companies are pushing ahead with pilots and platforms, even though their starting points vary widely. “Everyone’s engaging,” Rowan said. “Everyone’s got something going from an AI perspective. But the field still has very vast gaps in terms of level of maturity.”
Some are in the early stages, dipping their toes in AI chatbots. Others are moving quickly into agentic AI for revenue generation and growth. This disparity shapes how leaders interpret what they see in the market.
“You’re hearing snippets of stories across the board, and you’re not quite sure, ‘well, is that hype? Is that reality? Is that use case for me?’” he said. That dynamic has produced “hype fatigue” and a growing sense of FOMO, or fear of missing out, as companies try to read competitors’ moves.
AI vs. cloud and internet
The current AI boom is more consequential than earlier revolutions such as cloud computing and the rise of the internet, according to Rowan. The difference is the breadth of AI’s reach. Cloud was a targeted infrastructure shift. The web transformed communication. But generative and agentic AI are reshaping entire operating models, job categories and even the physical world through robotics.
“It is so ubiquitous to what’s happening across the economy, versus things like cloud and the internet,” Rowan said. “This is impacting all areas and levels of work, and I think that’s what makes it very different.”
The transformation is not limited to IT. “We’re hitting marketing, we’re hitting supply chain, we’re hitting IT, we’re hitting finance,” he said. “It’s all the elements of the business coming together.”
Software engineering illustrates the magnitude of the change. Coding assistants now generate nearly all of the work for some teams. “Leaders out there are saying that 90% of their code is being written around some of their tooling,” Rowan said. “You’ve seen a massive shift in two years.”
Image generation, multimodal agents and reasoning improvements are also improving rapidly. The pace is forcing organizations to rethink benchmarks for reliability. “You need new benchmarks around some of the things we’re looking at,” he said, noting that reliability is no longer optional. “You can’t have a guess at the next right answer in a supply chain.”
Another shift is the widening gap between consumer and enterprise AI adoption. Consumers see new capabilities instantly, thanks to continuous product updates, while enterprises must first navigate governance, risk, legacy systems, cost and technical debt.
“The consumer experience with AI is happening faster than the enterprise experience,” he said. This tension was absent in the development of cloud computing, which was revolutionary for companies but not consumers per se.
Accuracy improves, costs drop
For years, accuracy concerns − especially hallucinations − slowed enterprise adoption. But those fears are abating. “Accuracy is definitely improving. We’re seeing that statistically,” he said.
A larger shift is architectural: Companies are beginning to use ensembles of agents and models, then apply a ‘judge agent’ to select the best outcome. “Using that kind of architecture further refines the answer,” Rowan said.
Costs are improving just as quickly. Token prices are falling, and lightweight models that are less compute-intensive can provide strong performance for routine tasks. “We continue to see the cost per token continue to go down,” he said. Flash models “perform really well at a lower cost point” while bigger models handle heavier workloads.
Yet this flexibility introduces a governance challenge: Who decides which model to use, and when? “You don’t want to leave that to your users to navigate,” Rowan added.
Rowan said the AI talent picture is often misunderstood. Companies generally understand their data talent but underestimate skill gaps in AI modeling, agentic architectures and evaluation frameworks. “Enterprises have probably a good sense (of the data skills) in their organizations, but maybe not as much on the AI skills,” he said.
However, the tools themselves are quickly democratizing who can meaningfully contribute. Employees can now use applications that trigger agent workflows without being AI engineers themselves. Low-code and no-code interfaces, he added, will accelerate that trend.
At the other end of the spectrum, elite researchers developing frontier models will remain beyond the reach of most enterprises. “Those are fewer and far between,” he said. “Enterprises probably won’t have access to that talent.” Instead, companies tap their expertise through model APIs and partnerships.
Two AI futures
In mapping out the future, Rowan sees two plausible scenarios. The first is steady progress, with enterprises continuing to adopt at their own pace while the technology improves incrementally. “We’re just sort of moving along at the pace of the enterprise,” he said.
The second scenario is a break-the-curve moment triggered by a sudden leap in capability that pushes AI systems “beyond the expert level of experiences needed for a lot of tasks,” according to Rowan.
When that happens, adoption will become urgent. “The rush to really transform your enterprise is going to happen really fast,” he said. Companies that wait “are going to be very frustrated if they haven’t started internally, because it will be very hard to try to catch up in the future.”
He expects entirely new markets will emerge, citing agentic commerce as one example. In that world, “we’re not going to go to a store to buy goods. We might go there to experience the goods in a different way,” Rowan said.
If he had one piece of advice to executives, it would be to “use AI every day in your day-to-day life,” he said. “Be the champion in the enterprise for AI adoption. You’re not going to be sorry about it in the future.”
