You’ve probably heard some version of, “AI is changing everything,” in a sales pitch, on LinkedIn, or from a vendor hoping to win your business. What you may not have heard is how much pressure there is behind it, or how much it’s distorting the way companies make decisions right now.
Mitchell Hashimoto, a well-known software founder, put it bluntly earlier this year, writing that entire companies are “under heavy AI psychosis,” so caught up in it that a rational conversation about the technology has become nearly impossible. He’s far from alone. Consultants, engineers, and executives keep describing the same pattern: organizations so swept up in AI excitement that the calmer voices have learned to stay quiet out of fear for their jobs.
As a company that builds websites and applications for small businesses, we watch closely where this technology earns its keep and where it’s being oversold. In researching this piece, reading first-hand accounts, industry reporting, and talking with people who work inside these companies, one picture keeps repeating: the people in charge often have no real plan, and many privately doubt the strategy they’re publicly championing. Almost no one feels safe saying so.
This affects you even if you never buy a single “AI solution.” The vendors you depend on, the software your business runs on, and the partners you hire are all working inside this same environment. Understanding what’s actually happening helps you make steady, clear-headed decisions while the people around you chase a trend.
Here’s what the mania looks like from the inside, and how to keep your own judgment intact while it runs its course.
Most AI Projects Are Failing Quietly
Ask whether all this AI spending is actually paying off, and you’ll struggle to get a straight answer. Almost everyone involved has a reason to say that it is. Executives who admit a company overreached tend to lose their standing, employees who raise doubts get managed out, and vendors have every incentive to call a pilot a win.
The research that has cut through the noise is sobering. MIT’s NANDA initiative analyzed 300 public AI deployments and found that 95% of enterprise generative AI pilots delivered no measurable impact on the bottom line, while only about 5% produced a real return.
The failures usually have less to do with the technology than with how companies run projects. A business that already struggles to ship software on time doesn’t get better at it by bolting on a new, unpredictable tool. The AI inherits every normal way a project can go wrong, then adds a few of its own.
The version you’re most likely to run into is the chatbot. Internal ones tend to sit unused, because a chatbot can only answer from what a company has actually written down, and most companies’ documentation is thin. Customer-facing ones are the riskier bet, because now the failures happen in front of the people you’re trying to keep. I’ve yet to have a single satisfying interaction with one. Most talk me in circles and can’t connect me to a human who could actually fix the thing I called about.
One consultant writing about this described contacting a major automaker’s support line after his car broke down. A polite, natural-sounding AI system took his details and promised a callback. The call never came. Nothing registered as an error, so on paper the system looked like it had handled the request. What actually happened is that the company lost a repeat customer, who went and bought his next car elsewhere.
A similar thing happened to me. I ordered business cards from a printing company through their website, then needed to cancel, because we were updating our logo and I wanted the new one on the cards. The site had no phone number, only a chatbot, and it couldn’t touch my order or reach anyone who could. I ended up with a run of cards carrying the old logo, which I handed out just so they wouldn’t go to waste. I won’t use that company again.
A tool can post clean numbers while costing you the very thing those numbers were meant to protect. So when you hear that a company “boosted productivity with AI,” it’s worth asking what got measured, whether anyone tracked real usage, and whether the customer on the other end ever got what they needed.
Belief Became a Loyalty Test
In a lot of large companies, endorsing AI stopped being a business judgment and turned into a test of loyalty. Championing it is safe. Questioning it, even to ask what problem it’s meant to solve, can put your standing or your job at risk. That pressure shapes almost everything you hear coming out of these organizations.
Part of what it produces is performed belief. People declare that “AI is changing everything,” then can’t name a single thing at their company that actually changed. In one account, an executive rolled out an AI-centered strategy for a multibillion-dollar business shortly after admitting they had never once used an AI tool in their life.
The numbers suggest this is widespread. A survey of 1,200 tech workers and executives found that 79% of workers, and about 91% of executives, admitted to overstating how much they actually know about AI.
The same pressure produces fake usage. A survey of more than 1,000 US professionals found that 22% feel pushed to use AI in situations that make them uneasy, and 16% admit to pretending to use it when they aren’t. Engineers describe being scored on how much AI they consume, so they set the tools running in a loop to run up the numbers while they do the real work by hand.
None of this is harmless play-acting. When doubt is career-ending, the people who could tell you “this isn’t working” learn to stay quiet, and the only signal left is applause. Even some engineers who are broadly optimistic about AI have called the top-down mandates to adopt it bad strategy, for exactly this reason.
If you’ve felt behind for not rebuilding your business around AI, or a little foolish for being skeptical, it helps to know how much of the enthusiasm around you is pressure rather than results. The loudest confidence is often the least tested, and the people who would tell you otherwise have mostly learned to keep it to themselves.
Don’t Let the Demo Do Your Thinking
A polished demo is the most dangerous thing in the room, because it’s built to slip past the part of your brain that asks hard questions. The tool that looks like magic for two minutes on a screen is often the same tool that comes apart the moment it meets your real data and your real customers.
Take the feature nearly everyone wants right now: ask your data a question in plain English, like “what was our revenue last week,” and get an instant answer back. In a demo it looks flawless. On a real company’s messy data, it usually isn’t.
Research backs this up. Leading models that answer correctly around 90% of the time on clean, academic test data drop to roughly 20%, and in some tests into the single digits, once they’re pointed at real enterprise databases. Its a quiet failure. The tool hands back a number that looks perfectly reasonable and is simply wrong, with nothing to flag why. The AI app that dazzled on stage can be wrong more often than it’s right on your actual books.
The strange part is what a demo like that does to a room. One consulting team described showing this exact kind of AI feature to clients who were only lukewarm on their actual services. Every time, even after being told directly that the tool would not do what they wanted, the client wanted to buy it on the spot, brushing past millions of dollars in value the team could have delivered by ordinary means. They were unsettled enough that they pulled the demo from their lineup entirely. A responsible advisor doesn’t show off a tool they would never actually recommend.
The clearest warning sign is your own reaction. If a demo leaves you itching to buy immediately, sweeping aside the price, the caveats, and every other option on the table, treat that pull as a reason to slow down rather than sign. Ask what it does on a bad day, on your data, in front of the customers you can’t afford to lose.
A demo proves a tool can look impressive under perfect conditions. Your business runs under real ones.
No One Wants to Break Ranks First
Many of the executives making bold AI claims in public don’t believe them in private. The confidence you’re hearing is often fear wearing a steadier face.
Executives say as much themselves. A Cisco survey of 2,500 CEOs found that nearly two-thirds worry they’re underinvesting in AI, and Deloitte’s research describes leaders treating adoption as a business imperative driven by the fear of falling behind, with the hype pushing companies into premature spending. The engine is the fear of being left behind, not proof that any of it pays off.
So why doesn’t anyone with sense correct the record? Consultants who sit in these rooms describe a trap. A vendor can’t tell a big customer that their “100x productivity” claim is implausible without seeming to call that customer’s leadership foolish, and that’s a fast way to lose the contract. That vendor is also a customer of other vendors, who are stuck in the same bind. So the inflated claims travel up and across the market, and everyone keeps nodding, because whoever speaks first pays for it alone.
The doubt is there; it just runs silent. One 2026 survey found that confidence in an eventual payoff fell the further you got from the top job, from 62% among CEOs to 48% among executives outside the C-suite. The people closest to the real work are the least convinced, and the least free to say so.
You are not in that chain. You don’t answer to a board demanding AI theater, you have no press release to defend, and none of your contracts depend on praising a tool that doesn’t work. It leaves you free to judge AI on what it actually does for your business.
The people setting the tone can’t afford to stop and ask whether any of it works. Nothing stops you from asking, and your business runs on the answer.
In Part 2 we’ll get into what this means for your business and how to use AI without getting swept up in the frenzy.