When was the last time you spoke to a human sales representative or help desk?
Every company wants to tell you it’s “AI-powered” now. Here’s the part they leave out of the keynote: a lot of the time, the AI is making a mess, and a human somewhere is quietly cleaning it up, or increasingly, standing in a courtroom explaining why the chatbot promised something the company refuses to honor.
Welcome to the age of AI mediocrity. Not AI apocalypse. Not AI utopia. Just AI sloppiness, deployed at billion-dollar scale. Let’s take a tour.
Chapter One: Customer Service, But Worse
The great promise of AI customer service was speed and availability. The great reality, so far, is confident nonsense delivered with a smile.
The canonical case is Air Canada. A passenger named Jake Moffatt used the airline’s website chatbot while booking a flight after his grandmother died. The bot told him he could apply for a reduced bereavement fare retroactively, within 90 days of travel. That was wrong, since the policy explicitly excluded retroactive applications. When Moffatt asked for the refund the chatbot promised, Air Canada refused and then argued, in what the tribunal called a “remarkable submission,” that the chatbot was a separate legal entity responsible for its own actions. The British Columbia Civil Resolution Tribunal was not amused, found the airline guilty of negligent misrepresentation, and ordered it to pay damages.
Then there’s DPD, the parcel delivery company, whose chatbot responded to a frustrated customer by swearing, insulting its own employer, and composing poems about how terrible DPD was. The exchange went viral before the company disabled the feature.
And in a case that should be taught in every business ethics class, a Chevrolet dealership’s chatbot was manipulated into agreeing to sell a new Tahoe for one dollar, and declaring the offer legally binding. The dealership never had to honor it, but the lesson stands: if your bot can be talked into anything by a bored teenager with a prompt, it is not a customer service tool. It’s a liability with a text box.
Chapter Two: The Missing Orders
If AI is going to take our jobs, you’d hope it could at least take a drive-thru order. It cannot.
McDonald’s spent two years testing an AI drive-thru system with IBM at over 100 locations. The results were TikTok gold and operational garbage: orders of nine sweet teas nobody asked for, hundreds of dollars of McNuggets meals added mid-sentence, bacon on ice cream. In June 2024, McDonald’s quietly ended the partnership and pulled the plug. This is the company that basically perfected fast-food logistics, and voice AI still ate its lunch.
Cursor, an AI code editor beloved by developers, deployed a customer support bot named “Sam” that hallucinated a logout policy that didn’t exist, told wrongly logged-out users their frustration was “expected behavior,” and watched customers announce their cancellations on Reddit as a direct result.
The pattern: these systems don’t fail loudly. They fail confidently, which is worse, because confidence is what customers read as competence.
Chapter Three: Website Embarrassments
Google, the company that arguably started this entire era, has provided some of its most instructive failures.
In February 2024, Google Gemini’s image generator produced historically absurd outputs: Black Vikings, a female Pope, racially diverse Nazi-era German soldiers. Google paused the feature entirely and issued a public apology, admitting its tuning failed to account for cases that should clearly not show a range. Google, with effectively unlimited talent and compute, shipped a product that couldn’t reliably draw the Pope.
And this wasn’t an isolated oops. In 2023, a single inaccurate answer in Google Bard’s launch demo, where the chatbot claimed the James Webb Space Telescope took the first photos of exoplanets, which it didn’t, knocked roughly $100 billion off Alphabet’s market value in a day. One slide, one sentence, one hundred billion dollars.
Meanwhile, Google’s Gemini demo video was later revealed to be edited, with sped-up, pre-scripted responses presented as real-time interaction. The product wasn’t just wrong; the marketing of the product was misleading.
Chapter Four: Wrong Data, Right on the Slide
Which brings us to the executive floor, where AI’s mediocrity gets dressed up in a blazer.
The most famous example isn’t a slide but a filing: lawyers who used ChatGPT for legal research submitted entirely fabricated court cases, complete with fake quotes and citations, into a federal court filing. A federal judge responded with a standing order requiring attorneys to certify no part of filings was drafted by generative AI.
But investor-facing materials are catching the same disease. The SEC has charged companies with “AI washing,” making false or exaggerated AI claims in investor-facing communications, and fined firms for overstating their AI capabilities in the materials used to raise money. And how many startup decks now contain AI-generated market-size numbers, AI-generated competitor comparisons, and AI-generated charts that no human actually verified? The hallucination doesn’t stop being a hallucination because it moved into PowerPoint. If anything, it gets more expensive.
So What Do We Actually Learn?
Strip away the comedy and a consistent set of business lessons emerges:
- You own what your AI says. Air Canada learned that the hard way. A tribunal isn’t going to accept “the chatbot did it” as a defense. Your AI is your brand.
- Confidence is not accuracy. These systems are optimized to produce plausible-sounding output, and plausible is the enemy of verified when money, legal exposure, or health is involved.
- Human-in-the-loop is not a buzzword, it’s a business requirement. The failures above share one thing: no competent human was checking the output before it reached a customer, a court, or a market.
- Speed of deployment is a choice, not a virtue. Google could have tested Gemini’s image generator longer. McDonald’s could have piloted more carefully. The race to be “first with AI” is producing a lot of firsts to apologize.
- AI should draft; humans should decide. The companies getting value out of AI aren’t the ones removing humans from the loop. They’re the ones moving humans to the final checkpoint.
The age of mediocrity isn’t caused by AI being bad. It’s caused by management deciding that “good enough, probably” is an acceptable standard for things customers can see. The technology will improve. The question for this class, and for every business you’ll ever work in, is whether the judgment improves with it.
