Earlier this year, Apollo chief economist Torsten Slok made an observation about AI’s “productivity paradox” as data emerged that employees could potentially save an entire workday a week by deploying AI, but economic data showed a steep drop in productivity: “AI is everywhere except in the incoming macroeconomic data,” Slok wrote in a blog post, echoing the famous paradox observed by Nobel laureate Robert Solow in 1987 during the information technology boom. “Today, you don’t see AI in the employment data, productivity data, or inflation data.”
New data from Boston Consulting Group (BCG) not only confirms a similar phenomenon is happening in the workplace, but sheds light on why, providing a potential jump-start for businesses experiencing this productivity paradox. BCG’s 2026 Global AI at Work report, surveying nearly 12,000 frontline employees, found 42% of respondents reported eight hours of saved time, the equivalent of one workday a week, as a result of regular AI use, but 66% said they received limited to no guidance on what to do with the time they saved. Half said they’re not using that saved time for more strategic work.
According to David Martin, Global leader of BCG’s People & Organization practice, the productivity paradox, at least at the workplace level, comes down to a very human failure: leaders are not communicating clearly just why and how AI should be used in the office.
“Senior leaders are really struggling to articulate what the vision and strategy is on AI,” Martin told Fortune. “Consequently, it increases employee fear. It makes it harder for them to even understand what objectives they’re pushing for, and it trickles through to adoption, usage, and the like.”
The rise of tokenmaxxing
AI’s ability to make good on tech leaders’ lofty promises of productivity gains has come under fresh scrutiny as workplace adoption has continued to rise. For one, the technology has proven itself to be more expensive than human labor, in part a result of the enormous costs to run AI models, as well as token-based pricing models hiking up companies’ computing fees in a way that far outpaces improved output. Microsoft reportedly canceled much of its direct Claude code licenses, and Uber burned through its entire 2026 AI coding tools budget in the first four months of the year. Microsoft AI chief Mustafa Suleyman said as much earlier this week, telling Bloomberg that Anthropic is simply too expensive and the company is seeking alternatives.
“For my team, the cost of compute is far beyond the costs of the employees,” Bryan Catanzaro, vice president of applied deep learning at Nvidia, recently told Axios.
These surges in token usage driving up companies’ costs have come as companies push employees to find AI productivity gains. The Financial Times reported last month that Amazon employees were “tokenmaxxing,” in order to meet the company’s AI metrics. Other tech companies have employed similar techniques, including Meta, which set up an AI user leaderboard.
Gil Luria, head of technology research at brokerage D.A. Davidson, previously told Fortune the strategy is effective for increasing token usage, but little else.
“That doesn’t sound very healthy,” Luria said. “You get the behavior that you create the incentive for. So if you tell people they’ll succeed if they use a resource more, of course they’ll use it more.”
The era of tokenmaxxing is over
To BCG’s Martin, the lack of company instruction around AI beyond calls to use it and use it often is why the technology’s potential benefits have laid fallow.
“This whole tokenmaxxing thing is has probably run its course, and now it’s hitting their cost base in a pretty big way,” he said, echoing reporting by Fortune‘s Jeremy Kahn. “A lot of companies just gave AI to everyone, regardless of position, and I think now they’ll say, ‘Well, let’s be more thoughtful about who has access, and what is the business case? And are we delivering on it, ultimately? Then holding people accountable to meeting their targets, just like they would anything non-AI that they’ve been doing for the past 100 years.”
Indeed, the same tech companies enabling tokenmaxxing are now ending incentives around AI use. Last week, Amazon scrapped its internal tracking of AI use after employees deployed AI bots to compete useless tasks, according to the Financial Times.
“Please don’t use AI just for the sake of using AI,” Dave Treadwell, an Amazon senior vice-president, reportedly told staff.
At the Fortune COO Summit this week, Okta COO and President Eric Kelleher said he sees an enormous failure of imagination around AI adoption. “Everyone has the mandate [to adopt AI],” he said, but people are asking the wrong questions instead. “We have trained every manager in the world to think about one thing and that is, what’s their headcount? What’s the org chart look like? Who reports to who? How many layers do we have?”
Another summit attendee, Rakuten International COO Adrienne Down Coulson, wrote in a Fortune commentary this week that C-suites pushing for additional AI use are far behind their “superhuman” workers. “Here’s the uncomfortable truth,” she wrote: “AI can’t fix a broken C-suite running on an antiquated operating system.”
Martin said companies thinking about how to integrate AI into their entire enterprise operating model have seen success in increasing AI adoption and leveraging the technology to increase productivity, but another element of success with use of workplace AI comes from assuaging fears around the technology. Recent BCG research found that when workplaces treated AI agents like digital employees as opposed to tools, it increased employee fears around being displaced. This fear inhibits workplace sharing and encourages secret AI use, he said.
“In a world where people are fearful, they are actually saying, ‘Okay, how can I get a leg up on my peers, so that I can position myself to be highly valuable?’” he said. “And because of that, they’re a little bit reluctant to talk to their peers about what they’re doing, and that has huge implications on the organization’s ability to move more quickly.”
One balm in easing AI fears has been introducing comprehensive upskilling training to workers, Martin said. Workers who feel more empowered are more likely to share resources with others, making a company more nimble.
“A sharing culture is incredibly important,” he said. “But it’s not natural for fearful employees.”
