A Fortune commentary states that the current AI economy is dominated by two opposing sentiments: “this time is different” on one side, and “no one knows the answer” on the other. The article argues that this coexistence of optimism and uncertainty is the most defining feature of today’s AI boom.
Professor Ethan Mollick of the Wharton School at the University of Pennsylvania recently stated at an event at the New York Public Library that even AI labs, corporate executives, and industry influencers lack a ready-made framework to answer how AI will truly transform businesses. He said that anyone claiming to already have the “standard playbook” is not trustworthy.
Overall productivity gains remain limited.
The article cites Bank of America data stating that AI currently contributes only about 0.1% to annual overall economic productivity. This figure stands in stark contrast to market expectations for AI. In the same report, Bank of America still describes AI as a technology with a greater impact than electricity and the internet.
Goldman Sachs' research in March this year reached a similar conclusion. Its report stated that no significant correlation has yet been observed between AI and productivity gains at the overall economy level. However, in industries where AI applications are more concentrated, such as customer service and software, median productivity gains have reached up to 30%.
According to Bank of America’s estimates, AI can currently transform about 20% of workplace tasks, of which only 23% are cost-effective at this stage. Even if fully automated, the labor cost savings would amount to approximately 27%, while labor costs themselves make up roughly half of total costs. Based on this metric, the theoretical upper limit for current labor productivity gains is about 0.66%, and actual implementation would further reduce this due to friction and execution delays.
Internal corporate processes are slowing down implementation.
The article argues that the delayed full realization of AI benefits is less likely due to the technology itself and more due to organizational structure. Mollick notes that corporate IT departments are often where AI projects most easily stall—not because they oppose innovation, but because their inherent responsibilities lean toward risk control.
He also noted that the KPI system can limit room for experimentation. If a company demands from the outset that a project must deliver a 10% improvement, it often ends up selecting only minor tweaks to existing processes rather than driving replacements for the processes themselves. In other words, AI applications that truly transform the way work is done may not thrive within traditional performance frameworks.
Even AI companies are still exploring deployment methods.
The article also highlights a more telling phenomenon: many AI companies are building their own consulting and deployment teams to help clients integrate models into real-world business operations. Mollick believes this itself indicates that the industry has not yet developed a mature, replicable path for implementation.
If AI models have become powerful enough to reshape a large number of white-collar jobs, these companies should find it easier to answer the fundamental question of “how to deploy.” Yet in reality, even the most enthusiastic AI providers are still figuring out how to implement AI in enterprises.
The article argues that the core contradiction in today’s AI industry is not merely about valuation levels, but rather the misalignment between the pace of technological advancement and organizations’ ability to absorb it. While the market believes AI will bring profound changes, it lacks directly applicable implementation pathways—this tension will continue to shape the next phase of the AI economy.