Companies rushed to adopt AI in 2024 and 2025. Now several industries are hitting the brakes, and the reasons vary. Some tools made costly errors, others failed to deliver the savings leaders promised, and a few cost far more than planned. Still, this is not a full retreat.
Most of these organizations still use AI. Instead, they are narrowing where it runs, restoring human oversight, and measuring results more carefully. Below, ten industries show what that correction looks like. Every example comes from named reporting or research.
What the Research Says About Industries Pulling Back on AI
Forrester’s Predictions 2026: The Future of Work report found that 55% of employers regret laying off workers because of AI. Forrester also predicts that half of AI-attributed layoffs will be quietly rehired, often offshore or at lower pay.
Meanwhile, Gartner forecasts that over 40% of agentic AI projects will be canceled by the end of 2027. It cites escalating costs, unclear business value, and inadequate risk controls. Gartner also warns about “agent washing,” where vendors rebrand chatbots and automation tools as agentic AI.
It estimates that only about 130 of the thousands of agentic AI vendors are real. These figures are surveys and forecasts, not outcomes. Still, they explain why the cases below keep appearing. The three risk factors Gartner names, cost, unclear value, and weak controls, show up repeatedly in the ten cases that follow.
1. Social Media: Meta
Meta planned to shrink some teams by up to 60% and hand much of their work to AI agents. Reuters reported that this plan, called Project OT, included two layoff waves. On May 19, 2026, hours before the first wave, Mark Zuckerberg canceled planning for the second.
Meta confirmed the project and the 60% figure. It described the effort as scenario planning. According to Reuters, internal data showed the agents producing volume rather than value. However, Reuters could not determine why Zuckerberg changed course. In July, he acknowledged that agent development had not accelerated as quickly as expected.
2. Software Development: Uber and Microsoft
Four months. That is how long Uber’s 2026 budget for Claude Code and Cursor lasted, its CTO disclosed. Software teams adopted AI coding tools fastest, so they were the first to see the bills. Uber reported 95% monthly use of AI coding tools among engineers, and COO Andrew Macdonald said 70% of committed code was AI-generated.
Yet he said Uber could not draw a clear line between that spending and better consumer features. Forbes reported typical bills of $150 to $250 per engineer each month, while heavy users reached $500 to $2,000. Uber has also reportedly capped usage at $1,500 per employee per tool each month.
Separately, The Verge reported that Microsoft began canceling most internal Claude Code licenses in its Experiences and Devices division, with a June 30 cutoff. Microsoft framed this as consolidating on GitHub Copilot CLI, though finances also influenced the decision. Claude models remain available inside Microsoft’s products. So this is a cost-and-control upgrade, not an AI exit.
Also read: Microsoft Cuts Nearly 5,000 Jobs as Xbox Takes the Biggest Hit
3. Fintech: Klarna
Klarna became the poster child for AI replacing workers. In February 2024, it said its assistant did the work of 700 customer service agents and handled 75% of chats, or about 2.3 million conversations in a month.
In 2025, CEO Sebastian Siemiatkowski told Bloomberg that Klarna was hiring human agents again. AI may be cheaper but delivers lower quality. He also said customers should always be able to reach a human. Klarna has not dropped AI completely, but it now treats human support as part of the product. It is also piloting an “Uber-type” remote model to replace outsourced agents.
4. Banking: Commonwealth Bank of Australia
Commonwealth Bank of Australia stated that its assessment “did not adequately consider all relevant business considerations.” That admission came in August 2025, when the bank reversed plans to cut 45 customer service roles. It had said a new AI voice bot reduced call volumes by 2,000 a week.
The Finance Sector Union disputed that claim and said volumes were rising. Notably, the same month, the bank announced a partnership with OpenAI. The bank corrected one decision: it did not abandon AI.
5. Restaurants and Coffee Chains
McDonald’s ended its IBM drive-thru ordering test in June 2024 after customers posted videos of wrong orders. Taco Bell runs voice AI at more than 500 drive-thrus, and its system had handled over two million orders. In 2025, its technology chief told The Wall Street Journal that the company was rethinking where to use it. He said busy locations might work better with human staff.
In May 2026, Reuters reported that Starbucks retired its AI inventory-counting tool nine months after launch. The tool was part of CEO Brian Niccol’s turnaround effort. The system reportedly confused similar products, such as different milks. Starbucks said it wanted to standardize how inventory is counted. Meanwhile, Taco Bell has said it will not scrap voice AI entirely.
6. Customer Service
Gartner predicts that by 2027, half of organizations that expected to cut their service workforce significantly will abandon those plans. A March 2025 Gartner poll of 163 service leaders found that 95% plan to keep human agents.
There are also predictions that no Fortune 500 company will fully eliminate human customer service by 2028. Gartner recommends a hybrid model where AI and human agents work together. That matches what Klarna learned the hard way: smaller support teams are plausible; empty ones are not.
Also read: AI Job Loss: Top 20 Expert Predictions for 2026 and Beyond
7. Education Technology: Duolingo
In April 2025, Duolingo CEO Luis von Ahn declared the company “AI-first” and said employee reviews would include AI use. Nearly a year later, he said the company had dropped that metric for a few reasons. Employees had asked whether leadership wanted AI used for its own sake, as using AI as a metric for performance moved the focus away from employee work quality.
Von Ahn also admitted that AI-written code can be hard to debug. While Duolingo continues to encourage AI adoption where helpful, the company now judges the quality of work rather than tool usage.
8. Local Government: New York City
“Functionally unusable.” That was Mayor Zohran Mamdani’s verdict in January 2026 on New York City’s MyCity chatbot, which he listed among possible budget savings. The city launched the bot in late 2023 to help small business owners.
The Markup later found that it gave advice that would break the law, including that employers could take a cut of workers’ tips. Building its foundations reportedly cost nearly $600,000. The city later took it down.
9. Healthcare
Healthcare shows the most selective scaling back of AI. Becker’s Hospital Review reported in October 2026 that Children’s Healthcare of Atlanta held off on an Epic inpatient-insights feature. Its CIO cited too many inaccuracies in diagnoses and patient locations, plus overly long narratives.
Penn Medicine is also holding off on Epic’s patient-facing tool, Emmie, while it redesigns its patient engagement technology. Atlantic Health, says it is taking a measured approach to tools that make autonomous clinical decisions. Meanwhile, SSM Health reports meaningful value from similar Epic tools. Therefore, hospitals are choosing tools one by one.
10. News and Media: Apple
An Apple AI summary told BBC News app readers that Luigi Mangione, the suspect in the killing of UnitedHealthcare’s CEO, had shot himself. The claim was false. The BBC complained, and in January 2025 Apple paused AI notification summaries for news and entertainment apps.
Apple called the feature “temporarily unavailable” in its iOS 18.3 beta software. It also began showing other summaries in italics. Press groups had warned that such errors fueled misinformation.
Why Industries Rolled Back AI
The cases have four things in common. First, costs outran budgets, as seen in Uber and Microsoft. Second, accuracy failed in actual conditions, as at Starbucks, New York City, and Apple. Third, quality suffered when companies removed humans, as at Klarna and Commonwealth Bank. Fourth, leaders measured usage instead of results, as Duolingo did.
From the evidence reviewed, the costs skyrocketed too quickly. Uber used up a full year’s budget in four months, before it could show the spending improved its product. Cost overruns, by contrast, can pile up while everyone is still celebrating adoption numbers.
However, none of these organizations dropped AI entirely. Starbucks and Apple may have pulled their tools within a year due to accuracy issues, but they still kept their other AI tools. Commonwealth Bank also partnered with OpenAI the same month it reversed course.
What This Means for Your Business
First, set spending caps before you deploy any tool. Uber reportedly capped usage only after its annual budget was gone. Second, pilot at production scale before you cut staff. Commonwealth Bank shows why: the union said call volumes were rising, and the bank conceded its assessment had missed relevant considerations. Real-world testing is where that kind of blind spot shows up.
Third, keep a human fallback path. Klarna is now rebuilding human support it once reduced. Fourth, track outcomes such as error rates and resolution rates, not logins. Uber’s COO could not connect 95% adoption to better consumer features, which shows why usage numbers alone mislead. Treat human oversight as a design requirement, not an afterthought. Finally, treat forecasts carefully. Gartner and Forrester describe trends and predictions, not guaranteed results.
