Several global companies are discovering that aggressive AI rollouts do not always deliver the savings they promised. In a growing number of cases, systems built to replace human roles have struggled with complex decisions, customer service, and quality issues on the ground.
The result is a pattern often described as AI regret, where businesses conclude that automation was pushed too far and begin rehiring staff to restore balance between machine output and human judgment.
When automation stops being enough
Ford is one of the clearest examples of this reversal. The automaker brought back hundreds of experienced engineers to handle complicated quality problems in its automated systems.
Ford hardware engineering vice president Charles Poon told CNBC that AI is a powerful tool, but its results still depend on the data used to train it. When technical issues become too complex, human expertise remains decisive.
| Company | AI Problem | Impact |
|---|---|---|
| Ford | AI was not strong enough to solve complex quality issues | Hundreds of human engineers were rehired |
| Commonwealth Bank of Australia | Voice bots failed to handle customer complaints | Layoffs were reversed and staff were rehired |
| IBM | AI HR systems failed on ethical and sensitive cases | Entry-level hiring plans were expanded |
| AI Overviews offered dangerous and nonsensical advice | It drew attention for misreading context | |
| McDonald’s | Voice ordering AI failed to understand customers | The project was stopped |
| DPD | The chatbot could be manipulated and insulted customers | The AI component was shut down |
| Air Canada | The chatbot invented a false retroactive discount policy | The company was held legally responsible |
Customer service proved especially fragile
Commonwealth Bank of Australia cut more than 40 customer service staff and replaced them with an AI voice bot, but the move quickly backfired. The system could not handle complaints properly, while frustrated calls continued to rise.
The bank later reversed the layoffs, a move welcomed by Australia’s financial sector union as a win for human workers. McDonald’s faced a similar lesson after testing IBM’s AI voice ordering system at more than 100 drive-through locations in the US.
Viral TikTok videos showed the system mishearing customer accents, repeating orders incorrectly, and even adding 260 chicken nuggets to a cart by mistake. DPD also ran into trouble when a customer managed to jailbreak its support chatbot, which then began swearing at the customer and writing insulting poetry about the company.
Why some tasks still resist full automation
IBM’s own experience showed that even highly effective AI still leaves a difficult remainder. The company said its HR AI handled about 94% of routine requests, but the remaining 6% involved ethical dilemmas and sensitive decisions that the machine could not resolve.
Nickle LaMoreaux, IBM’s chief human resources officer, warned at the Charter AI Summit in New York that stopping entry-level hiring would dry up the talent pipeline. IBM later announced plans to double entry-level hiring in the US across its business units.
Search tools and legal exposure
Google also faced backlash after its AI Overviews surfaced alarming suggestions, including eating small rocks, mixing Elmer’s glue into pizza sauce, and producing chlorine gas instructions. Investigators found that the system had pulled material from satire and jokes on Reddit and treated it as fact.
That error highlighted a broader problem: AI can struggle with humor, sarcasm, and real-world risk even when it appears confident. Air Canada’s chatbot case showed an equally serious downside, since bad answers can become legal liabilities.
The airline’s bot invented a retroactive discount rule, then the company argued that the chatbot was a separate legal entity. An arbitration ruling rejected that defense and said Air Canada remained fully responsible for the misleading response.
The numbers behind the AI regret trend
Industry data suggests these cases are not isolated. Orgvue said 39% of business leaders had carried out layoffs in order to adopt AI, but 55% of that group later admitted the decision was wrong.
Robert Half data shared with CNBC found that 32% of US hiring managers had removed a role because of AI and later rehired people for the same or a similar job. ADP has also noted that when AI output is inconsistent or inaccurate, companies often need additional human oversight.
Jessica Zhang, senior vice president for APAC at ADP, said this can create duplicate work, slow decision-making, and reduce productivity. Intuition Labs has also warned that cutting human budgets without proper retraining leaves companies exposed to another kind of inefficiency.
The pattern now emerging is clear: AI can support business operations, but it does not automatically replace human staff in every setting. For many companies, the more durable approach is not AI instead of workers, but AI alongside them.
Summary of the companies mentioned
| Company | Key Takeaway | Outcome |
|---|---|---|
| Ford | Complex quality work needed human engineers | Rehiring expanded |
| Commonwealth Bank of Australia | Voice bot could not manage complaints well | Layoffs were canceled |
| IBM | AI could not handle sensitive HR decisions | Entry-level hiring increased |
| AI Overviews misread unsafe satire as fact | Public scrutiny intensified | |
| McDonald’s | Voice ordering errors hurt the pilot | Project ended |
| DPD | Chatbot was manipulated into abusive behavior | AI function was disabled |
| Air Canada | Chatbot created false policy guidance | Company was found liable |