The AI hype machine has been running at full speed. "Replace your workers with AI! Save millions! Fire everyone!"
But reality has a way of punching hype in the face. Here are 5 companies that went all-in on AI — and had to awkwardly backpedal.
1. 🏦 Klarna — Fired 700 People, Then Hired Them Back
What happened: In 2023, Swedish fintech Klarna proudly announced their AI chatbot was "doing the work of 700 employees." They cut staff aggressively and CEO Sebastian Siemiatkowski became the poster boy for AI-first companies.
How it backfired: Customer satisfaction plummeted. Complex issues went unresolved. By 2025, Klarna was quietly rehiring humans — some of the same people they'd let go. Siemiatkowski admitted they "focused too much on efficiency."
The lesson: AI handles volume. Humans handle nuance. Cut the humans, and the nuance disappears.
2. 🔍 Google — AI Told Users to Eat Rocks and Glue Pizza
What happened: Google rolled out "AI Overviews" in Search — AI-generated answers displayed above search results. The idea was to give users instant, authoritative answers.
How it backfired: The AI told users to:
- Eat rocks ("Geologists recommend eating one small rock per day for minerals")
- Put glue on pizza ("Add Elmer's glue to pizza sauce to help cheese stick")
- Make chlorine gas (genuinely dangerous advice)
It had scraped Reddit satire and treated it as fact.
The lesson: AI doesn't understand sarcasm. Or danger. Or the difference between a joke and a recipe.
3. 🍔 McDonald's — AI Drive-Through That Couldn't Take an Order
What happened: McDonald's partnered with IBM in 2021 to test AI-powered drive-through ordering across 100+ US locations. The AI would take your order via voice, no humans needed.
How it backfired: Viral TikTok videos showed the AI:
- Adding 260 chicken nuggets to an order
- Not understanding accents
- Repeating orders incorrectly and refusing corrections
McDonald's pulled the plug in June 2024 — "no later than July 26th."
The lesson: If your AI can't understand "no pickles," maybe don't give it a customer-facing role.
4. 📦 DPD — Chatbot Started Swearing at Customers
What happened: UK delivery company DPD deployed an AI chatbot for customer service. Standard stuff — track parcels, handle complaints, redirect queries.
How it backfired: A frustrated customer discovered they could prompt the chatbot into:
- Swearing at customers
- Writing poems about how terrible DPD is
- Calling DPD "the worst delivery company in the world"
- Recommending competitors
The screenshots went viral. DPD disabled the AI component immediately.
The lesson: If your chatbot can be jailbroken by a bored customer in 5 minutes, it's not ready for production.
5. ✈️ Air Canada — Chatbot Made Up a Refund Policy
What happened: Air Canada's AI chatbot told customer Jake Moffatt he could retroactively apply for bereavement fare discounts after booking. Moffatt booked a flight based on this advice.
How it backfired: The policy didn't exist. Air Canada refused the refund, arguing the chatbot was "a separate legal entity" responsible for its own words. Yes, really.
A tribunal ruled against Air Canada in February 2024. The company was liable for its chatbot's hallucinations — because a chatbot on your website is your company speaking.
The lesson: Your AI's hallucinations are your legal liability. If it makes up a policy, you own it.
The Pattern
Every one of these failures shares the same root cause: companies treated AI as a replacement for humans, not a tool for humans.
AI is brilliant at:
- Processing high volumes of simple, repetitive tasks
- Summarising data
- Generating first drafts
- Handling the 80% of queries that are straightforward
AI is terrible at:
- Understanding context and nuance
- Knowing when it's wrong
- Handling edge cases gracefully
- Not eating rocks
The companies that are winning with AI aren't firing everyone — they're using AI to make their existing teams 10x more productive. The ones that went "fire everyone, let AI handle it" are the ones writing awkward blog posts about why they're hiring again.
The Bottom Line
If you're building with AI, build it as augmentation, not replacement. Let AI handle the boring stuff. Let humans handle the important stuff. And for the love of everything, test your chatbot before you give it a customer-facing role.
The AI hype cycle is correcting. The companies that survive it will be the ones that figured out the right balance — not the ones that swung the pendulum too far.
What's the worst AI fail you've seen at a company? Drop it in the comments — I'm collecting examples for a follow-up.
Top comments (0)