Have you noticed the deafening silence? For all the chest-thumping and nine-figure investment announcements about artificial intelligence, there’s a conspicuous lack of victory laps. Every conference stage features a CEO proclaiming AI will revolutionise their business, yet if you pull a mid-level manager aside for a quiet pint, you’ll hear a very different story. A story of stalled projects, bewildering complexity, and budgets spiralling into the abyss. This isn’t just a few isolated incidents; it’s a systemic, industry-wide phenomenon. Welcome to the great AI cover-up.
The central paradox of our time is this: AI investment continues to soar, even as evidence mounts that most projects are failing. The dirty little secret is that a fog of AI implementation secrecy has descended upon the corporate world. Companies are terrified of admitting they’ve bet the farm on a technology they don’t fully understand and can’t seem to get working at scale. They’re caught in a trap of their own making, and admitting failure feels more dangerous than throwing good money after bad. But why? What forces are compelling supposedly rational executives to keep quiet and keep spending?
The Honey Trap of Vendor Lock-in
Let’s start with the most insidious problem: vendor lock-in. When a company decides to “get serious about AI,” the first call is often to one of the big cloud players—Amazon, Microsoft, or Google. They arrive with slick presentations, promising a seamless, all-in-one solution. It feels safe. It feels easy. Your board has heard of these companies. It’s a decision that’s easy to justify.
But this convenience comes at a steep price. Once you build your entire AI strategy on a single vendor’s proprietary tools and infrastructure, you’re stuck. Moving your data becomes a Herculean task. Retraining your models on a different platform is prohibitively expensive. Your engineers are now specialists in one specific ecosystem, making them less flexible. You’ve essentially handed the keys to your kingdom to another company, and they can—and will—raise the rent whenever they please. It’s no wonder executives don’t want to talk about it. Admitting to vendor lock-in is admitting to a massive strategic blunder right at the start of the journey.
Think of it like deciding to build a house, but instead of hiring an architect and a builder, you buy a plot of land inside a theme park that owns the only hardware shop, employs all the plumbers, and dictates what colour you’re allowed to paint your door. Sure, it was quick to get started, but now you can’t change a thing without their permission and their price tag. This lack of flexibility is precisely what kills genuine innovation. You end up optimising for the vendor’s toolkit, not for your actual business needs.
Drowning in AI’s Technical Debt
Running parallel to vendor lock-in is the equally corrosive problem of technical debt. In the frantic rush to show some AI progress, teams are encouraged to cut corners. They build quick-and-dirty pilots, hardcode solutions that should be flexible, and choose expediency over robust engineering. Each of these shortcuts is like taking out a high-interest loan. You get the cash (a flashy demo) upfront, but the repayments (bugs, maintenance nightmares, and an inability to scale) will eventually cripple you.
This isn’t a new concept in software, but AI adds a dangerous new dimension. The complexity of machine learning models means that the “interest” on technical debt accrues at an alarming rate. A poorly documented data pipeline or an unversioned model from six months ago can render an entire project useless. As companies rush to stitch together various AI services, they are creating monstrously complex systems that no single person understands.
When these pilot projects inevitably collapse under the weight of their own debt, what happens? They aren’t dissected in post-mortems to learn valuable lessons. Instead, they are quietly shelved. The failure is blamed on a “bad data set” or “shifting business priorities.” The underlying structural problems—the debt—are ignored, and the organisation moves on to the next shiny AI object, destined to repeat the same mistakes. This is a core component of the AI implementation secrecy; it’s easier to hide the mess than to undertake the costly and embarrassing clean-up.
An Economist’s Reality Check on the AI Gold Rush
While the tech world loves to preach disruption at lightning speed, economists offer a more sobering perspective. As Martha Gimbel of the Yale Budget Lab pointed out in a recent MIT Technology Review report, “It would be historically shocking if a technology had had an impact as quickly as people thought that this one was going to.” This is a crucial point. Genuine technological revolutions, from electricity to the internet, take decades to fully permeate the economy and reshape business processes.
The current AI hype cycle has compressed this timeline into a matter of months, creating wildly unrealistic expectations. We’re measuring a marathon in 100-metre sprints. The data supports this slower reality. A recent analysis cited by the MIT Technology Review found that a staggering 95% of generative AI pilots were failing to meet expectations. That isn’t a sign that AI is useless; it’s a sign that we are terrible at implementing it. Companies are trying to bolt a jet engine onto a horse-drawn cart and then acting surprised when it doesn’t fly. The smart money is on firms that understand this is a long game, not a “get rich quick” scheme.
The Rise of Innovation Theater
If 95% of pilots are failing, why is every company still talking about AI? This brings us to the most cynical part of this whole charade: innovation theater. This is the art of looking innovative without doing the hard work of actual innovation. It involves grand announcements, billion-dollar investment pledges, and the creation of “AI Centres of Excellence” that produce more press releases than products.
Innovation theater is a performance for investors, the board, and the media. It’s about managing perception. Coca-Cola, for example, announced a splashy $1.1 billion deal with Microsoft for AI, yet details on widespread adoption or tangible impact remain scarce. It looks good on a slide deck, but what does it actually do? This performance creates a vicious cycle. When one company announces a big AI spend, its competitors feel pressured to do the same, regardless of whether they have a coherent strategy. The result is a massive, industry-wide bonfire of cash, with very little actual heat being generated. The pressure to keep up appearances is a powerful motive for maintaining AI implementation secrecy. No one wants to be the first to admit the emperor has no clothes.
Learning from the Fallen: McDonald’s and Taco Bell’s AI Retreat
While many failures are hidden, some happen right in front of our eyes. Take the fast-food industry’s brief and calamitous affair with AI-powered drive-throughs. Both McDonald’s and Taco Bell have recently abandoned high-profile AI ordering systems after they proved to be, to put it politely, a complete mess. Bacon was added to ice cream cones. Hundreds of chicken nuggets were added to a single order. Customers grew frustrated.
These public failures are incredibly valuable. They show us that interfacing with the messy, unpredictable real world is still fantastically difficult for AI. A chatbot in a controlled digital environment is one thing; an AI trying to understand a muffled order over a cheap speaker next to a noisy road is another entirely. Instead of hiding these results, companies should be studying them. The lesson isn’t “AI is bad.” The lesson is that we have grossly underestimated the complexity of the problems we’re trying to solve. But in a culture that punishes failure, these lessons are often ignored in favour of finding a “safer” project that looks good on paper.
The Smart Money is on Patience: Lessons from Klarna and Coca-Cola
So, is anyone getting this right? The answer seems to be a quiet “yes,” but it doesn’t look like the hype. Look at Klarna, the “buy now, pay later” firm. After initially boasting that AI could do the work of 700 customer service agents, the company quietly changed its tune. Their new mantra? “AI gives us speed. Talent gives us empathy.” They realised AI is a tool to augment their human workforce, not replace it. It’s a subtle but profound strategic shift away from techno-optimism towards pragmatic reality.
This reflects a more mature approach: strategic patience. It’s about resisting the pressure to launch half-baked AI features and instead focusing on building a solid foundation. It means investing in data quality, training your people, and carefully choosing small, well-defined problems where AI can have a real impact. It’s less glamorous, but infinitely more likely to succeed. It’s about escaping the cycle of innovation theater and playing the long game.
Time to End the Secrecy
The current state of AI adoption is unsustainable. The combination of vendor lock-in, crippling technical debt, and the performative circus of innovation theater is leading countless companies towards a cliff edge. The pervasive AI implementation secrecy is preventing the entire industry from learning from its mistakes.
It’s time for a reckoning. Boards and investors need to start asking tougher questions. Instead of “How much are we spending on AI?”, the question should be “What is our strategy for avoiding vendor lock-in?” or “How are we managing the technical debt from our AI experiments?”.
For business leaders, the path forward requires courage. It requires admitting that you don’t have all the answers and that the initial rush was perhaps misguided. It means shifting focus from flashy demos to foundational work: cleaning up data, upskilling staff, and fostering a culture where small, contained failures are treated as learning opportunities, not career-ending mistakes. The companies that navigate this transition successfully will be the ones that own the next decade. The rest will be left explaining to their shareholders what happened to all that money.
What do you think? Are you seeing this AI implementation secrecy in your own organisation? How do we break the cycle of hype and failure?
