An Autopsy of AI Hype: What Harvard Business Review Got Wrong
In this blog
Most people spend their Christmas break eating leftovers or watching football. I spent mine in the kitchen with a razor knife, a high-speed scanner and a copy of Harvard Business Review's (HBR) 2019 anthology: HBR's 10 Must Reads on AI, Analytics, and the New Machine Age. I physically sliced the binding off the book, fed the loose pages into a my trusty scanner and loaded the text into my all-time fave AI tools to perform a forensic audit on the recent past.
It's a strange hobby, I admit, but it revealed something fascinating.
Back in the pre-pandemic economic landscape of 2019, the corporate elite had a specific vision of the future. It was a world of "Human+AI" symbiosis in retail, blockchain supply chains and drones buzzing overhead. This vision was built on the shaky foundation of the Zero Interest Rate Policy (ZIRP) era, where capital was cheap and "growth at all costs" was the law of the land. Six years later, in the cold financial light of 2025, that vision has crumbled. By comparing those earlier predictions to the just-released State of AI in Business 2025 report from MIT NANDA, a starkly different reality emerges.
Here is my autopsy of the predictions that failed and the hard business facts regarding what is actually working today.
The 2019 autopsy: Why the experts were wrong
To understand what works with AI today, we must first dissect the well-intentioned but misguided predictions of the recent past. The 2019 Harvard Business Review anthology barely mentioned large language models (LLMs). Instead, it bet the house on hardware, complex logistics and "human-in-the-loop" services. It was a future that never arrived.
The "Stitch Fix" fallacy: When humans are too expensive
In 2019, the "Human + AI" model was pitched as the future of retail. The idea was that algorithms would do the math and humans would add the empathy, creating a virtuous flywheel. The company Stitch Fix is a great example of this trend, which gained popularity as a data-driven, personalized online styling service that blended human stylists with technology to help users discover styles that fit their taste, size and budget.
The reality is that the "human" layer of this approach turned out to be a massive variable cost that scaled linearly with revenue. Unlike software, you have to pay a person to check every box. The verdict is clear: Stitch Fix stock collapsed roughly 95% from its highs*. You simply can't have software margins with a service-based workforce.
*It should be noted that there are signs of a Stitch Fix stock rebound, thanks in part to the use of AI.
Alexa: The $10 billion timer
Harvard Business Review predicted that marketing would transform into a "battle for AI assistants' attention," where brands would pay a premium "slotting fee" to get prime placement in Amazon's algorithm. The reality is that consumers rejected "voice shopping" because it is cognitively heavy and slow. We use Alexa for timers, weather and alarms — utilities, not commerce. The verdict? Amazon's "Worldwide Digital" unit, which houses Alexa, has reportedly hemorrhaged over $10 billion in annual operating losses.
Blockchain: The ledger to nowhere
Maersk's TradeLens platform was once cited as the foundational technology for a new economy, promising a "single source of truth" for global shipping. It shut down completely in 2023. The "trustless" technology failed because of fundamental human trust issues. Competitors simply refused to join a platform owned by a powerful rival, rendering the entire concept useless.
Augmented Reality (AR): The hardware graveyard
"Why Every Organization Needs an Augmented Reality Strategy," one Harvard Business Review article declared, predicting that smart glasses would become the new interface for industrial work. But in reality, Microsoft discontinued the HoloLens 2 and Google ended its Glass Enterprise edition. The verdict is that workers refused to wear heavy, hot headsets for eight-hour shifts. The smartphone remained the dominant, more practical interface. The common thread in these failures is a focus on complex hardware and logistics, while the real revolution was brewing in software.
The 2025 reality check: What actually works
If the ZIRP-era predictions were so wrong, where are we now? According to the new State of AI in Business 2025 report from MIT NANDA, we are in the middle of a massive "GenAI Divide." The "New Machine Age" has arrived, but it came via software code, not drones. Understanding the following five "Hard Facts" from the report is critical for staying on the right side of that divide.
Hard fact 1: The 95% failure rate
Despite $30-40 billion in enterprise investment, 95% of organizations are getting zero return from their AI initiatives. Most companies are stuck in "pilot purgatory." They have high adoption of generic tools but see low transformation of actual business processes. The lesson is simple: If an AI initiative doesn't touch the P&L, it is just a toy.
Hard fact 2: The "shadow AI" economy is real
While official enterprise initiatives stall, employees are already crossing the GenAI divide without you. Only 40% of companies have purchased an official LLM subscription, but 90% of employees report using personal AI tools for work. This happens because many enterprise tools are brittle, over-engineered "wrappers" that don't actually help people work faster. Your people know what good AI looks like and they will use their own tools if the ones you provide are inadequate.
Hard fact 3: The "learning gap" kills pilots
Enterprise tools often fail because they have amnesia. The report finds that 90% of users still prefer humans for complex tasks because current AI tools "break in edge cases" and "don't learn from feedback." As one user put it, "It's useful the first week, but then it just repeats the same mistakes." The future belongs to Agentic AI— systems with persistent memory that can learn from their mistakes over time.
Hard fact 4: Stop ignoring the back office
A massive "Investment Bias" exists in corporate AI, with roughly 70% of GenAI budgets flowing to Sales and Marketing. While these projects promise vague "productivity gains," the real money is in the unsexy work of Operations, Finance and Procurement. Companies are seeing $2-10 million in annual savings by automating BPO (Business Process Outsourcing) tasks, a tangible return that marketing projects often lack.
Hard fact 5: Buy, don't build (seriously)
In perhaps the report's most controversial finding for IT teams, strategic partnerships ("buying") are twice as likely to succeed as internal "build" initiatives. Internal builds fail more often because most companies underestimate the complexity of maintaining an AI system that actually learns. The winning strategy is to treat AI startups like BPO providers, not just software vendors, holding them accountable for outcomes, not just uptime. These facts paint a clear picture of an industry shifting from speculative bets to a focus on tangible, process-level value.
The call to action: Get practical with automation
As I fed the last shredded pages of that 2019 book into the scanner, I realized the authors weren't wrong about the destination — a frictionless, automated economy. They were just wrong about the vehicle. They thought the vehicle was "magic" like blockchains, drones and holograms. But the vehicle is actually just "mechanics" — data, APIs and integration.
This brings us to the most important takeaway for 2025: All AI is automation, but not all automation is AI. Our WWT Research report on Automation Priorities for 2025 shows that the most successful companies are not just buying AI models; they are automating the messy, boring layers of their infrastructure first.
Two of the most glaring failures from 2019 are perfect examples of this principle in action.
The "boring" fix for the "shadow AI" crisis
Consider the "shadow AI" crisis from the MIT report, where 90% of employees are using unauthorized tools. The "AI" solution might be to buy a multi-million-dollar AI Governance Platform that few will use. The "automation" solution is much simpler: better IT Asset Management (ITAM). If your software asset management were automated, "shadow AI" would not be a terrifying security risk; it would be a visible data stream. You could see the usage patterns and make strategic decisions, like securing an enterprise license for a popular tool, instead of banning it and forcing users underground.
The foundational fix for the AR/VR failure
The same logic applies to the failure of enterprise AR. Companies bought HoloLens headsets but had no data to show inside them. They couldn't augment reality because they hadn't indexed reality first. If they had automated their Hardware Asset Management and digitized their physical inventory, the AR platform would have had a "Digital Twin" foundation to stand on.
The final verdict: From the new machine age to the new practical age
The MIT report proves what pragmatic leaders already know: The only AI that delivers value is "customized," "integrated" and "process-specific." In other words, it's not magic — it's mechanics. Successful AI isn't a disembodied brain you buy, it's a nervous system you build by deeply integrating it with your existing data, APIs and workflows. Translated from analyst-speak, that means AI works when it is treated as automation, not magic. Go ahead and shred your old playbooks. The future isn't about the New Machine Age. It's about the New Practical Age. Stop chasing the hype. Start automating the foundation.
Practical AI in action
As we've learned, the journey to AI success begins with automation. Automation is the essential foundation that enables AI to function effectively, as seen in Fortune 500 companies that prioritize automating core processes before deploying AI technologies. This approach ensures data integrity, scalability and operational agility, allowing AI to deliver intelligent decision-making and transformative value.
What does this look like in the real world? In the banking sector, for example, AI and automation are rapidly transforming workplace services, enhancing productivity and operational resilience. By focusing on pragmatic use cases and strong governance, banks are realizing real improvements in employee productivity and satisfaction.
Moreover, the integration of AI with automation is crucial for operationalizing AI insights. Without automation, AI-driven analysis remains disconnected from business processes, resulting in impressive demos that never reach production. By connecting AI to automation platforms, organizations can automatically trigger workflows and scale AI-driven automation across practical use cases, turning AI insights into actionable business results.
A key component of this transformation is IT Asset Management (ITAM). Effective ITAM practices ensure that organizations have a clear view of their software and hardware assets, enabling better decision-making and risk management. By automating ITAM processes, companies can gain visibility into asset usage patterns, optimize resource allocation, and make strategic decisions about securing enterprise licenses for popular tools. This not only mitigates the risks associated with "Shadow AI" but also transforms ITAM from a reactive process into a proactive strategy that supports AI and automation initiatives.
In conclusion, the path to AI excellence is paved with automation and robust ITAM practices. By embedding automation throughout the enterprise and leveraging ITAM to manage assets effectively, organizations create the structure and consistency AI needs to perform effectively. This practical approach to AI ensures that businesses not only survive but thrive in the new practical age.