Leonard Lee is the Executive Analyst and Founder of neXt Curve, a tech industry research and advisory firm.
Seven years ago, I wrote a piece called “Want To Get Value Out of IoT? Stop Talking About IoT!” At the time, the tech hype of the moment was IoT. Much like today’s generative AI boom, pundits predicted IoT would usher in the next industrial revolution—only instead of AI factories, it would be smart factories.
So what makes AI different from IoT? Not much.
Like IoT, AI is not a singular technology. It is a concept that encompasses a wide range of scientific research, frameworks, methodologies and foundational technologies used to build intelligent systems and applications. Most of the time, AI manifests as a capability or feature embedded in a broader product or service. Take computer vision, for example, as it appears in advanced driving assistance systems (ADAS) in modern vehicles.
When the IoT hype didn’t deliver on its hockey-stick growth forecasts, evangelists coined a new term—AIoT. This concept muddied the waters even further. Even adding AI couldn’t lift IoT out of its long winter.
Today, generative AI (GenAI) risks following the same trajectory.
We’re seeing massive investment in GPUs, hyperscale data centers, and GenAI startups—many of which statistically won’t survive. But none of that is AI. It’s infrastructure—supercomputing environments optimized or a narrow set of tasks, such as large-scale model training.
Over the past couple of years, many enterprises have launched proof-of-concept (POC) or pilot initiatives to explore GenAI’s much-hyped potential.
Those early explorations often involved experimentation with prompt engineering—a practice that quickly proved fleeting as models evolved from simple LLMs into today’s more advanced reasoning models. We’ve also seen global diversification, with companies like China’s DeepSeek introducing their own GenAI models and unique innovation.
Show Us The Money
More than two years into the GenAI hype cycle, tangible value remains elusive for many stakeholders—from major vendors and VCs to the enterprises footing the bill.
The dilemma GenAI faces is similar to past hyped technology cycles like IoT or blockchain. Boards and executive teams push “AI transformation” programs that promise big returns but are often based on theoretical benefits no one has yet realized. As even vendors and consultants will admit, enterprise AI is still in its infancy—and we’re all still learning.
Adoption remains tepid. Many initiatives are stuck in pilot purgatory, slowed by the mismatch between deterministic expectations and a probabilistic technology.
The emergence of retrieval-augmented generation (RAG) hasn’t been the breakthrough hoped for. Persistent challenges—security, safety and the fundamental limitations of RAG stacks—are difficult to overcome. A vector database, for example, is not a relational database. And as I predicted years ago, the lack of embedding-level role-based access control (RBAC) leaves a glaring security gap in most GenAI architectures.
Add to that the persistent issues of hallucinations, model drift and collapse—the costs of building secure, sustainable GenAI applications become clear.
Best Practices To Shift The Conversation To Real Value
If organizations want to shorten the path to value, they need to shift their focus. These best practices learned from previous hype cycles can help organizations avoid common pitfalls and wasted investment.
1. Stop Talking About AI
It’s counterintuitive, but helpful. Most people don’t really understand AI—inside or outside your organization. Conversations often become mired in vague jargon and misaligned assumptions. This confusion frequently leads to programs that are poorly scoped, misdirected, or ultimately ineffective.
2. Focus On The Problem To Be Solved
Yes, it’s a cliche—but it’s also critical. Start with a clear, specific business problem. Then identify whether AI-related capabilities can augment existing tools or enable new ones. Like the IoT community eventually realized, articulating value in business terms—rather than technical abstraction—leads to better outcomes.
3. Being First Doesn’t Equal Leadership
FOMO is real. But if you want to save money, let your competitors go first and learn what NOT to do from their mistakes. There’s nothing wrong with waiting to see what works. In fact, it might be the smartest move you can make right now.
Even cloud leaders are taking a cautious view. As Amazon.com President and CEO Andy Jassy said, “It’s actually quite difficult to build a really good generative AI application.” I couldn’t agree more.
4. Let The Technology Experts Be The Technology Experts
Trustworthy AI starts with trusted AI experts. And the best ones aren’t just fluent in what GenAI can do—but also in what it can’t. Successful enterprise AI requires deep, full-stack expertise, especially across areas like security, safety and AI lifecycle operations.
Choose your experts wisely. They are your foundation for GenAI success.
Stop Imagining A Magic Pill
The sooner enterprises stop chasing the idea of AI as a magic solution, the sooner they can build real, meaningful capabilities. That starts with clear business goals, realistic expectations and teams equipped to navigate the hard problems—not just the hype.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
