Copy article

Tokenmaxxing: Workers Are Gaming the AI Metric Companies Use to Gauge Who's Replaceable

ended 22. May 2026

Tech-savvy workers have realised AI usage leaderboards don't just track adoption. They map which workers AI can replace, and staff have figured that out.

Major tech companies now track token consumption on internal leaderboards linked to performance reviews. The stated purpose is measuring productivity. The observable function is building a dataset of who does what, how often, and how easily an AI could do it instead. Workers who use AI least become visible. Workers who use it most teach it their job.

So employees are flooding the metric with noise, firing AI tools at pointless tasks to inflate their scores. It looks like enthusiasm. It functions as self-defence. If the leaderboard is a replacement index, tokenmaxxing is data poisoning.

Meanwhile the same workers still curate output that's rejected 70–90% of the time, supervise agents that delete production environments, and watch employers burn through annual AI budgets by April. 

We'd like your views:

  • If token consumption leaderboards generate data on task type, frequency, and complexity, what stops that dataset becoming a redundancy shortlist?
  • Workers are corrupting adoption metrics that feed into $700 billion in industry capex decisions. If the demand signal is partly performative, how reliable are the infrastructure forecasts built on it?
  • When AI output is rejected 70–90% of the time, is "curation" a new unpaid job, and should it be compensated as a distinct valuable skill?
  • Tokenmaxxing is Goodhart's Law as self-preservation. Is it the first organised labour response to AI displacement that doesn't involve a union?

3 responses from the Newspage community

Copy all

Star Quote
Copy

Colette Mason
AI Ethics Consultant at Clever Clogs AI
Amazon mandated that 80% of devs use AI weekly, tracked token consumption on internal leaderboards, and within 3 months AI-assisted code changes contributed to outages that cost 6.3 million orders in a single day. The same company now tracks who uses AI least, while 35% of managers say replacing staff with AI is a good idea, up from 23% last year. Workers flood the leaderboard with junk tasks to corrupt the signal.

Every junk token has a real cost. Claude Code subscriptions consume £6–£10 in compute for every £1 they pay, which means tokenmaxxing doesn't just game the employer, it bleeds the model providers too.

That compute runs on infrastructure on track to drink 5 bn cubic metres of water a year by 2027 and consume more electricity than Japan. This is burned so leaderboards can tell a manager who to replace, while the same compute could be used for medical research, agriculture optimisation, or other benefits for humanity.

The AI arms race is catching more in the crossfire.
Copy

Editor
Katrina Young
Futurist | Chief Technology Officer CTO/CDO Ethical AI & Digital Transformation Strategist at KYC Digital
Tokenmaxxing is not really about workers gaming AI metrics. It is about workers recognising those metrics may eventually be used to judge their value. The problem is that token usage does not measure actual contribution. It measures activity. It cannot see judgement, decision-making, experience, or the time spent correcting poor AI outputs before work reaches a client or production environment. Once employees believe AI usage scores affect visibility or job security, behaviour changes. People start optimising for the metric instead of the outcome. The data becomes unreliable. That creates a bigger issue for employers. If leadership starts using AI adoption data to shape hiring, restructuring, or performance decisions, they may be relying on a distorted signal. A leaderboard can track usage. It cannot measure human value.
Copy

Editor
Mitali Deypurkaystha
AI Governance Strategist & CMO at AISensa
One in four British workers already fear AI could cost them their job, according to workplace experts, Acas, so it is hardly surprising that “tokenmaxxing” is emerging as the first quiet rebellion against algorithmic management. These AI leaderboards do not just measure productivity; they risk becoming de facto redundancy maps that show employers which tasks are repetitive, predictable and easiest to automate. Once workers suspect every prompt helps train a system to replace them, the metric stops measuring efficiency and starts triggering self-preservation behaviour, with staff flooding dashboards with performative AI usage to poison the signal. Tokenmaxxing is essentially Goodhart’s Law in office wear: the moment AI adoption scores become tied to survival, employees stop optimising for output and start optimising for visibility. What looks like enthusiastic AI uptake may actually be a growing crisis of trust inside the modern workplace.