The tokenmaxxing era is over. What comes next is a people problem
Most organizations spent the past two years telling employees AI was the future. Some built leaderboards. Others tied AI usage to performance expectations. The message was clear: use it, use it often, use it visibly. Employees responded with tokenmaxxing, maximizing AI token consumption to climb internal leaderboards and signal their value. Now the bills are arriving, and the message is changing.
Uber burned through its entire 2026 AI tools budget by April. ServiceNow exhausted its annual token allocation before mid-year. Amazon shut down its internal AI leaderboard after employees gamed it to inflate their scores. Meta, meanwhile, sent an internal memo warning roughly 6,000 staffers that the company was on track to spend billions on internal AI use in 2026 alone, and that token budgets, allocations, and controls were coming.
The tokenmaxxing era, it seems, is over. What replaces it is still being worked out, and the people best positioned to shape it are HR leaders, if they move now.
What a token actually is
Before HR leaders can engage in the budget conversation, they need to understand what they're being asked to manage.
Julia Dhar, Managing Director and Partner at Boston Consulting Group and North America leader of BCG's People and Organization practice, explains the concept simply.
"A good way for non-technical folks to think about it is as a unit of data used by a large language model to process and generate a response. You can also loosely think about it as the work done, the number of tokens being the amount of work done by an AI model," Dhar said.
AI vendors have been shifting from flat-fee subscriptions to usage-based billing, charging for every prompt, every agent task, every automated workflow. With organizational AI adoption hitting 88% in 2025, according to Stanford's 2026 AI Index Report, the scale of that consumption was always going to add up.
"To generalize, in most large companies there wasn't consistent messaging to people that tokens were not free and that they cost the organization money, and there was some limit on the appropriate amount to be used, as well as guidance on the right type of tasks to use them on," Dhar said. "That isn't necessarily a critique, because I also understand the objective that leaders had at the time to encourage use in the organization."
Kate Connor, Chief Operating Officer of Moxie Communications Group, has seen this dynamic from the inside. Her team put per-team token budgets in place after recognizing that untracked usage wasn't producing proportional value. The lesson she draws from watching organizations, including some that ran internal AI leaderboards, is that the incentive structure drives the behavior.
"Nobody should be using AI tools just for the sake of using them," Meta's CTO Andrew Bosworth wrote in an April memo to staff. "Token usage alone is not a measure of impact of any kind." The memo followed reports that Meta employees had burned through 73.7 trillion tokens in a single 30-day period while competing on an internal leaderboard called Claudeonomics.
A change challenge, not a technology one
The organizations most exposed are those that encouraged heavy AI use without ever setting clear expectations about cost or purpose.
"We spent a lot of time collectively, but especially from leaders, telling people to get on the bus in relation to AI, and it maybe wasn't very clear where that bus was going," she said. "This has brought that to a head."
That reversal is its own communications challenge. Employees who were told AI adoption was a core expectation are now being told to pull back, and in a workforce already uncertain about what AI means for their jobs, mixed signals compound the anxiety.
"This is number one a change challenge, not a technology challenge," Dhar said.
Research for Dhar's book How Change Really Works: Seven Science-Based Principles for Transforming Your Organization drew on interviews with more than 6,000 executives and employees and found that the number one emotion people feel in relation to change is curiosity, not resistance. But the number two emotion is anxiety. In big transformations, the job of HR leaders, Dhar argues, is to keep the organization oriented toward curiosity and away from anxiety.
"People who are anxious can do a lot of things. It's just a really hard way to live," she said.
How to measure AI value instead of AI volume
If token consumption is the wrong metric, what should organizations be tracking instead? Dhar's answer starts with auditing what's actually happening before reaching for new controls.
"The first thing that any HR leader, any technology leader, and ideally the two together grappling with this right now can do is examine past usage as an indicator of what individuals in the organization are using these tools for," she said.
That audit, she argues, will reveal three patterns. Some employees are building deeply valuable, domain-specific tools that represent exactly what human-AI collaboration should look like. Others are duplicating effort, building the same custom tool independently across hundreds of teams. And others are simply inefficient, running expensive queries that could be structured more cheaply without any loss of output.
Duplicated effort is one of the most common patterns. Thousands of employees independently building the same tool doesn't just create governance headaches; it burns through tokens at scale with no organizational oversight.
"The classic HR example that CHROs share with me all the time is that in large organizations, hundreds or thousands of people have built their own tool to assist them in the process of writing performance reviews," Dhar said.
Connor's approach at Moxie starts with the budget. "There needs to be strict budgeting around token usage," she said. "There is a responsibility at the leadership level to put budgets in place on a per team basis." None of it, she added, sits with any single function. "I don't think any of these decisions truly sits with just HR or just tech or just finance. It's truly a cross-functional leadership transformation that needs to happen."
Dhar's framework for moving from adoption to accountability centers on what she calls a judgment-based policy, something between a blanket ban and unlimited access. The question she'd have organizations ask isn't "is this expensive?" but "can you defend this use?"
"Do you have a really clear value creation thesis? Do you expect that there will be a return on investment for what you are doing?" she said. "What degree of confidence do you have that this option is superior to the alternatives?"
HR's seat at the table
Dhar points to a structural problem underneath the token cost crisis. In many organizations, the AI agenda is being driven by the CEO, the chief AI officer, or the chief technology officer. HR has been an afterthought, if it's been included at all.
"One of the concerns I hear from CHROs and chief people officers about themselves and their own role in these discussions is a feeling that they don't have a seat at the executive table," she said.
The token budget moment is a way in. The decisions about what work AI does, how usage is governed, what behavior gets incentivized, and how the transition is communicated to employees aren't technology decisions. They're people decisions. And the organizations that treat them as such are more likely to get the outcomes they want.
As HRD America has reported on how AI governance is becoming a core HR mandate, the function that manages how humans and machines collaborate is increasingly the function being handed governance responsibility for the tools themselves.
"Now is a perfect moment for much more prominence for the HR team and HR leaders to seize this part of the agenda," Dhar said. "If there is a feeling that people have that they don't have a seat at the table, and it's accurate, and you've been looking for a way in, the way in which humans behave, the direction they get from their manager, their understanding of how to steward a company's resources really effectively, that is the way in."
