A recent MIT Media Lab study draws a sobering picture of the current state of enterprise AI. According to “The State of AI in Business 2025,” 95% of generative AI pilots fail to deliver measurable business value.

The problem isn’t the technology. It’s how companies use it.

Too many organizations are chasing trends instead of strategies. They launch AI projects not because they’ve identified a clear problem, but because everyone else seems to be doing it. AI has become a checkbox, a sign of innovation rather than a vehicle for impact.

‍

The Mirage of Quick Wins

The MIT researchers found that roughly 70% of corporate AI budgets are spent on sales and marketing pilots. These are easy to visualize: chatbots that talk to customers, tools that generate copy, systems that reply automatically.

But visibility isn’t value.

These projects often create noise, not results: chatbots that frustrate users, tone-deaf emails, content that feels automated rather than authentic. The real gains, the study shows, emerge in less glamorous places; in back-office automation, procurement, finance, and operations.

It’s there, deep in the structure of how companies actually work, that AI can generate true efficiency. But only when data, processes, and business logic are already aligned. Otherwise, as Forbes contributor Andrea Hill wrote, “Automating a flawed process only helps you do the wrong thing faster.”

‍

The Hype Gap: What Karpathy Gets Right

In his recent interview with Dwarkesh Patel, AI researcher Andrej Karpathy, formerly of OpenAI and Tesla, called out the over-optimism surrounding so-called Agentic AI.

Karpathy doesn’t see an “Age of Agents” arriving anytime soon. Instead, he describes it as the “Decade of Agents.” The current models, he says, “just don’t work” as autonomous systems. They lack real memory, reliable reasoning, multimodal understanding, and the capacity to operate independently in complex environments.

“I feel like the industry is making too big of a jump,” Karpathy explains. “It’s trying to pretend like this is amazing and it’s not. It’s slop.”

His critique is not pessimism. It’s realism. AI is powerful, but it’s not magic. Today’s models are, in his words, “autocomplete engines”  brilliant at prediction, limited in cognition.

‍

Why Companies Keep Failing

Both Karpathy and MIT point to the same underlying flaw: misalignment between technology and business reality.

Many enterprises operate in silos, strategy lives in PowerPoint, marketing in one system, operations in another. When AI is layered on top of this fragmentation, it amplifies chaos instead of reducing it.

MIT’s study also shows that externally partnered AI deployments succeed twice as often as purely internal builds (67% versus 33%). The reason is experience. Internal teams know the business; external partners know how to integrate, scale, and sustain technology across use cases. The most effective transformations happen when these two perspectives work together.

‍

Kumai’s Approach: Building Intelligent Systems, Not Just AI Projects

At kumai, we’re an AI consulting company that helps businesses turn ambition into impact.
We don’t build AI for the sake of AI, we build intelligent systems that solve real problems and create measurable results.

AI should not be the centerpiece of every solution, but when used intentionally, it can transform how companies operate. That’s why every kumai engagement starts with business intelligence, data integration, and process clarity the foundation on which sustainable AI value is built.

Every project begins with a readiness and data audit, an honest assessment of data quality, reporting structures, and information flows. Without that foundation, no algorithm can be trusted.
Once the fundamentals are solid, we align processes, KPIs, and ownership before introducing AI components designed to amplify what already works.

At kumai, AI isn’t decoration, it’s infrastructure.
We bring together data, technology, and strategy to help companies deploy AI with purpose, discipline, and real business impact

‍

From Pilots to Performance

The line between a pilot that fades and a system that transforms is integration.

AI can’t live beside your operations, it has to live inside them. When embedded directly into ERP, CRM, or process layers, it becomes part of decision-making, not an isolated experiment. That’s where real ROI begins: in the seamless movement between humans, data, and systems.

‍

How Companies Can Get It Right

The path to real AI impact isn’t complex — it’s disciplined.

1. Start with strategy. Define what success looks like before building anything.
2. Fix your data. Without clean, reliable data, there’s no intelligence — artificial or otherwise.
3. Include your people. Culture and communication determine adoption more than code.
4. Blend expertise. Combine business understanding with external implementation experience.
5. Think in systems. Integration turns pilots into performance

‍

A Measured Future for AI

Both the MIT study and Karpathy’s reflections remind us: AI isn’t the enemy — impatience is.

The winners of the next decade will not be those who deploy AI the fastest, but those who do it with purpose, data discipline, and operational maturity.

At kumai, we see AI as a force multiplier — but only when it stands on the shoulders of solid Business Intelligence and clear processes. We don’t chase hype. We build clarity.