But optimism may outweigh reality. MIT, The Genai divide: A new report from the AI state of Business 2025 draws calm pictures. As first reported by Fortune, the study reveals that 95% of businesses trying to integrate generated AI have failed. Only 5% of companies have achieved meaningful revenue acceleration.
The findings raise concerns that the AI gold rush could be inflated with dangerous bubbles.
Expectations meet overwhelming performance
For many years, Wall Street has taken AI banks and provided a historic boost to labor productivity. However, MIT researchers found little evidence that AI was moving the needle in that direction. Previous tests have shown that even the most advanced AI products only successfully complete around 30% of assigned office tasks. By April 2025, only 24% of real-world jobs were able to finish by April 2025, as it was exaggerated as “AI agents” (are touted as autonomous digital workers).
This gap between hype and performance has already forced businesses to rethink their strategy. A Gartner survey of 163 executives found that half had abandoned plans to dramatically cut customer service staff by 2027. “The human touch remains irreplaceable in many interactions,” says Kathy Ross, senior director of customer service and support at Gartner.
Workers are skeptical and businesses are retreating
Employees seem to share that sentiment. A survey of GOTO and workplace intelligence found that 62% of workers believe AI is “a massive exaggeration.” Many IT managers acknowledge that organizations do not have a formal AI adoption strategy, and security and integration challenges reach the list of barriers. The discrepancy between promises and delivery has already caused a public setback. Klarna, a Fintech company, cut nearly a quarter of its workforce in 2024 in anticipation of AIPERED’s future, and launched a recruitment drive to reverse courses earlier this year to rehire staff. Technology critic Ed Zitron took the mood straight away. AI “agents” often sound like intelligent assistants, but it is equivalent to “automating cards,” which requires important programming efforts from companies.
This gap between AI hype and performance has already forced businesses to rethink their strategy.
A new kind of bias: ai that chooses yourself over humans
Adding another wrinkle, researchers published in Proceedings of the National Academy of Sciences (PNAS) recently discovered an astonishing trend they call “AI-AI bias.” Larger language models such as GPT-4 and Meta’s Llama 3.1 consistently favored content created by other AIs over human-created materials, across product ads, academic abstracts, and even film reviews.
Research co-author Jan Kulveit warned that such biases could reshape economic opportunities and there is a risk that humans will be systematically sidelined. “Being human in the economy where AI agents live will suck,” he said in X. If another AI appears to be assessed, we advised people to perform the work through AI tools before submitting it.
This creates a troublesome picture. Not only does AI systems struggle to provide promised productivity gains, they may also be strengthening their own control at the expense of human contributions.
Bubble waiting for it to burst?
Financial interests are enormous. Analysts have long predicted that AI could add more than $6 trillion to the global economy by 2030. The major tech companies alone are hoping to earn another $600 billion in annual revenue.
Every year, AI is shortfalls, increasing pressure on future productivity, and increasing fear that the bubble is unsustainable. As MoneyWeek said, with a lot of money riding on AI, only a complete upheaval of existing systems would justify the investment frenzy.
The AI industry is currently at a crossroads. Breakthroughs remain possible, but MIT research and related research highlight that the actual impact is far behind investors’ expectations. If productivity doesn’t surge immediately, the “AI miracle” could prove to be a mirage.
For businesses and workers, the coming years will reveal whether AI is truly a revolutionary force or the latest bubble bulging by Silicon Valley optimism.
Altman’s Trillion Dollar Warning
Even as MIT researchers note the growing expectations, Openai CEO Sam Altman doubles the disruptive potential of AI. Speaking in San Francisco, he predicted that ChatGpt would be able to quickly handle more everyday conversations than humans have with each other. At the same time, he admitted to admitting to the crazy bear all the traits of the bubble that needed “trillions of dollars” in its infrastructure, leaving winners and losers alike. Whether AI is proving a durable revolution or another dot com style crash, Altman argues that the long-term economic return will be “huge.”
Verification tax will be added to the issue
The MIT findings are also consistent with concerns raised by industry leaders about what some people call “verification tax.” Tanmai Gopal, CEO of PromptQL, told Forbes that the biggest barrier is not the raw computing power, but the model’s “confident and wrong” tendency. The promised productivity gains often evaporate, as employees need to spend extra time double checking their output.
Researchers say this constant need for verification explains why many AI deployments have stalled and never expanded in pilot projects. In an industry where accuracy is key, even a single high confidence error can outweigh multiple successes and further erode trust and ROI.
