Important notice - We are aware of a scam where people are impersonating CMD Recruitment to offer jobs via WhatsApp & Telegram. This is a scam, CMD Recruitment will never contact new candidates via these methods. Any legitmate offer from CMD Recruitment would be made via a telephone call.
July 6, 2026
By Dan Barfoot, Operations Manager at CMD Recruitment | Updated 6 July 2026
Dan Barfoot is Operations Manager at CMD Recruitment, a recruitment agency working with employers and candidates across Wiltshire, Bath, Swindon and the wider South West, spanning HR, finance, administration, sales, engineering and technical disciplines. He works directly with employers on hiring strategy and people management, including the performance-measurement and workplace technology questions covered in this piece.
You’ve probably seen the term floating around LinkedIn, or in a headline about Meta, Disney or Uber. Tokenmaxxing is a workplace trend where employees consume massive amounts of AI compute, tokens, in the jargon, not because the work demands it, but to signal productivity or climb an internal AI-usage leaderboard. For HR, it’s a warning against rewarding superficial activity metrics instead of genuine, measurable business value.
The mechanics are straightforward. A company rolls out Claude, Copilot or a similar tool, wants to see adoption, and builds a dashboard to track it. Someone adds a leaderboard. Within weeks, employees are running AI agents unattended overnight and firing off pointless queries, anything that pushes their token count up, because the number on the leaderboard has become the thing that gets noticed. Not the work itself.
Let’s be precise, because the term gets used loosely online. Tokenmaxxing describes employees consuming large volumes of AI tokens specifically to signal productivity or climb an internal AI-usage leaderboard, rather than to produce genuinely useful work (Built In, 2026). It is not a story about artificial intelligence so much as a story about what happens when a business rewards the wrong metric.
The clearest example is Meta’s. In early 2026, a Meta employee built an internal dashboard nicknamed “Claudeonomics”, ranking the company’s 85,000-plus staff by AI token consumption and handing out titles like “Token Legend” and “Cache Wizard” to the heaviest users. Total usage on the dashboard topped 60 trillion tokens in a single 30-day window, and the top-ranked individual averaged 281 billion tokens on their own. Meta shut the leaderboard down within two days of it becoming public, after data from it leaked outside the company (Fortune, 2026).
Meta wasn’t alone. Disney gave close to 5,000 product and technology staff access to an internal “AI Adoption Dashboard” tracking use of Claude and Cursor, including a leaderboard-style view of top users, one employee reportedly invoked Claude around 460,000 times over nine working days. Amazon and JPMorgan have reportedly run similar internal leaderboards to encourage AI adoption (HR Executive, 2026). None of this is really about the AI. It’s about what happens when a business decides it wants to see “AI adoption”, picks the easiest number available to prove it, then makes that number visible, competitive and reward-linked.
If this feels familiar, it should. Long before AI leaderboards, employers built dashboards to prove employees were working. Hubstaff, Time Doctor, ActivTrak and Teramind, dubbed “bossware”, saw enormous growth from 2020 onward as remote work took hold; ActivTrak’s usage rose more than 150% and Teramind’s by roughly 800% within months of the first lockdowns, as employers scrambled to monitor newly remote staff (Business of Business, 2020). Microsoft’s own Productivity Score, launched the same year, let managers see individual employees’ email and chat activity, until a fierce backlash forced Microsoft to strip out the individual-level data in December 2021 and rebuild the tool around aggregated, anonymised metrics (NPI Financial, 2023).
It’s the same anxiety, in different clothes, that drove employers to push for a return to office a few years earlier: if you can’t see someone working, you look for a proxy that lets you believe you can. Swap “keystrokes” for “tokens” and the underlying flaw is identical. We work with employers across Wiltshire, Bath and the South West who are having exactly this conversation right now, not about mouse-jigglers any more, but about whether their new AI adoption dashboard is repeating the same trust-versus-surveillance trade-off we’ve seen play out before.
Tokens measure AI consumption. They do not measure AI-derived business value, and treating the two as interchangeable is where the trouble starts (CIO, 2026). An employee who uses a handful of well-targeted prompts to automate a two-hour task looks, on a token-usage dashboard, identical to one who leaves an agent running overnight purely to inflate their count.
Uber found this out the expensive way. The company burned through its entire 2026 AI budget in four months, largely on Claude Code, after ranking engineers on an internal leaderboard based on their usage; individual engineers were reportedly spending between $500 and $2,000 a month each on tokens before the company imposed a $1,500 monthly cap (Forbes, 2026). Uber’s COO, Andrew Macdonald, later said the connection between all that AI spend and any actual improvement for customers “is not there yet” (Fortune, 2026).
That’s the sharpest possible illustration of a point worth making about any metric: tracking a proxy for productivity is like measuring a chef’s performance by how many times they stir a pot. If the number can go up without the thing you actually care about improving, you haven’t built a performance metric. You’ve built a target for people to hit by whatever means available.
Tokenmaxxing didn’t stay an internal joke for long. Meta’s leaderboard only became public knowledge because data from it leaked outside the company, at which point Fortune and a wave of tech and HR commentary picked it up within days, and “tokenmaxxing” went from Slack-channel in-joke to a term HR publications like SHRM were writing explainers about within weeks (SHRM, 2026). That speed is itself instructive: assume any internal measurement system you build will eventually be visible externally, to journalists, to candidates researching your company, and to your own staff comparing notes with peers elsewhere.
This matters for employee retention strategies in 2025 and beyond. If your organisation is seen, rightly or wrongly, as a place that prizes visible AI activity over judgement and real output, you’ll struggle to attract and retain talent. Reputations built on leaderboard screenshots travel fast, and rarely in the direction employers want.
So what’s the alternative? IBM has put a name to it: valuemaxxing, optimising for the business outcome an AI-assisted task actually produces, rather than the volume of AI activity used to produce it (IBM, 2026). It’s a rebrand of an old idea. Best Buy tried something similar back in 2005 with its Results-Only Work Environment, judging staff purely on output rather than hours or visible activity, and by several accounts it worked: University of Minnesota research linked it to better sleep and lower turnover, though Best Buy quietly wound the programme down in 2013 as it retrenched (CNN Business, 2013), a reminder that these shifts need sustained management commitment, not just a launch announcement.
Frameworks like OKRs, the goal-setting system Andy Grove built at Intel in the 1970s and John Doerr later brought to Google, offer a structured way to define and track outcomes that actually matter, forcing managers to define what success looks like before the AI, or the employee, starts working (What Matters, 2024).
Trust-based management requires effort. It demands that managers hold regular, meaningful conversations about priorities, blockers and outcomes. It requires psychological safety in the workplace, employees need to feel able to say “I didn’t need much AI for this one” without it counting against them. None of that is achieved by a leaderboard.
If you’re an HR professional recognising this problem, here are some practical starting points.
First, audit anywhere AI usage volume shows up in a performance review, goal-setting conversation or manager report. SHRM’s own guidance is blunt on this point: once usage is visible and comparable, employees will optimise for it, whatever the intention behind the dashboard was. If your approach to hybrid work performance tracking, or increasingly AI-adoption tracking, relies on a number that’s easy to game, you already know the answer.
Second, separate adoption from value, deliberately. Redesign your metrics around business results, revenue generated, projects completed, hours genuinely saved on a defined process, verified rather than self-reported. These are harder to game because they reflect actual value, not activity.
Third, watch your budget as closely as your culture. Uber’s experience shows a leaderboard doesn’t just distort behaviour, it can distort your finance model entirely; token-based pricing does not behave like the software licence costs most finance teams are used to modelling. That is a conversation for HR and finance to have together, not one to leave to IT alone.
And finally, HR digital transformation means keeping pace with a genuinely new category of workplace risk, not assuming the AI-adoption metrics a vendor hands you are automatically the right ones to reward. We see the same pattern across the HR, finance and technical roles we recruit for throughout Wiltshire and the South West: employers who ask “what did this actually achieve?” before “how much did we use it?” end up with teams that trust the process, and stick around.
This isn’t a one-sided conversation. Employees chasing a leaderboard position should think about the longer-term trade-off. Racking up token usage doesn’t build a skill anyone will pay for later, knowing when and how to use AI well does. Relationships with managers become hollow if they’re built on a number rather than trust, and when a restructure or redundancy round arrives, as they inevitably do, the employees who can point to real, demonstrable outcomes are the ones who survive it, not the ones who topped an internal dashboard.
There’s a legitimate grievance underneath this trend too. Being told that “AI-driven impact” is now a core expectation, without a clear definition of what that means, is genuinely disorientating for employees trying to do a good job. The answer isn’t to fake the metric. It’s to ask, clearly and directly, how your contribution is actually being assessed, and to push for an answer that isn’t just a number on a screen.
The workplace is changing quickly, and AI is a genuine part of that shift. Both employers and employees have a responsibility to make sure that change rewards real value, not the cleverest way of gaming whatever happens to be easiest to measure.
Whether you are looking to refine your people management strategy, adapt your performance metrics for the AI era, or find your next career move in a forward-thinking workplace, navigating these shifts requires the right approach.
At CMD Recruitment, we connect forward-thinking employers with top talent across Wiltshire, Bath, Swindon, and the wider South West. We’re always happy to discuss how to align your hiring strategy and people management with authentic, measurable business value.
Get in touch with the team at CMD Recruitment today to find out how we can support your business goals or help you take the next step in your career.
Back to Blog
