Last Updated: February 2026
Meta just made it official. Starting in 2026, every Meta employee will be graded on their "AI-driven impact." Google has been quietly tracking how many extra productive hours its engineers gain from AI tools. A new wave of workplace surveillance software is surging in demand, all designed to answer one question: how much AI is each employee actually using?
Welcome to the age of tokens as a productivity metric. And it is both a breakthrough in workforce measurement and a minefield of bad incentives waiting to explode.
What Are Tokens and Why Do Companies Care About Them?
Tokens are the fundamental unit of AI interaction. Every time an employee uses ChatGPT, Copilot, or any large language model, the system processes their input and output in tokens (roughly 3/4 of a word each). The total tokens consumed across an organization represents, in theory, the volume of AI-assisted work being done. Companies are starting to track this number the way they once tracked email volume or lines of code: as a proxy for engagement and output.
The logic is straightforward. If your company is spending six or seven figures annually on AI tools, leadership wants to know the tools are actually being used. Token consumption per employee, per team, per department gives a clean, quantifiable signal of adoption. And in a world where AI fluency is becoming a core competency, that signal matters.
Which Companies Are Already Tracking AI Usage?
The trend is further along than most people realize. Meta is the highest-profile example, making "AI-driven impact" a formal component of performance reviews in 2026. But they are not alone.
- Google has been measuring the increase in productive hours per week generated by engineers using AI tools. CEO Sundar Pichai publicly reported a 10% productivity boost from AI adoption across engineering teams.
- Microsoft is measuring employee AI tool usage across multiple functions, tracking adoption patterns at both team and individual levels.
- GitHub CEO Thomas Dohmke sent an internal memo about tracking AI coding assistant usage across Microsoft, signaling that Copilot adoption is now a management priority.
- BCG found that 77% of leaders and managers use generative AI several times per week, but frontline employee adoption has stalled at 51%, creating pressure to close that gap through measurement.
Beyond the tech giants, an entire ecosystem of monitoring tools has emerged. Worklytics, ActivTrak, and Hubstaff all report sharp demand increases for AI usage tracking features. These platforms can identify which AI tools employees use, how frequently, and critically, who is not using AI at all.
Why Measuring Tokens Could Be a Terrible Idea
Here is where it gets dangerous. Measuring token consumption as a productivity metric has the same fundamental flaw as measuring developers by lines of code. More tokens does not mean better work. It might mean worse work.
A study by METR (Model Evaluation and Threat Research) found something counterintuitive: experienced open-source developers using AI tools were actually 19% slower than those working without AI assistance. The researchers suggested that context switching, prompt engineering, and verifying AI outputs created overhead that negated the speed benefits.
Consider what optimizing for token consumption actually incentivizes:
- Verbose prompting over efficient prompting. An employee who writes a detailed 500-word prompt and gets a mediocre answer consumes more tokens than one who writes a precise 50-word prompt and gets exactly what they need.
- AI busywork. If your review depends on demonstrating AI usage, the rational response is to route everything through AI, even tasks where it adds no value.
- Gaming the metric. Just as developers once padded code with unnecessary lines, employees will find ways to inflate token counts without improving outcomes.
- Punishing expertise. Senior employees who can solve problems from experience faster than an AI can generate and they can verify a response get penalized for efficiency.
Microsoft learned this lesson the hard way with their "Productivity Score" feature in Microsoft 365, which tracked individual employee activity (emails sent, meetings attended, documents edited). The backlash was severe enough that they had to anonymize the data and reframe the entire feature.
What Should Companies Actually Measure Instead?
The right approach is not to measure AI usage. It is to measure outcomes that AI usage should improve. The distinction matters enormously.
Outcome-based metrics that work:
- Time to completion. Did AI help this team finish projects faster? Measure cycle times before and after AI tool deployment.
- Quality indicators. Are there fewer errors, fewer revision cycles, higher client satisfaction scores?
- Capacity expansion. Can the same team now handle more work? More clients? More complex projects?
- Innovation output. Are employees producing more creative solutions, exploring more options, prototyping faster?
- Revenue per employee. The ultimate productivity metric. If AI is working, this number goes up.
Adoption metrics that make sense (as inputs, not KPIs):
- Tool activation rates. What percentage of employees have used the AI tools at least once this month? This identifies access and training gaps.
- Use case diversity. Are employees finding new applications, or just using AI for the same one task?
- Self-reported value. Anonymous surveys asking "Did AI save you time this week?" can surface genuine adoption patterns without surveillance anxiety.
The Real Risk: Creating an AI Surveillance Culture
Workplace sociologist Tracy Brower, who researches organizational behavior, noted that employers rarely make monitoring apparent, and employees often miss or forget internal communications about tracking. "It's like the terms and conditions of apps. Most people just click accept," she told Business Insider.
The deeper problem is cultural. If employees feel surveilled on their AI usage, you get compliance without conviction. People will use the tools to check the box, not to genuinely transform how they work. And the employees most likely to push back are often the experienced, senior staff whose judgment about when AI helps (and when it does not) is exactly what organizations need.
The companies that will win the AI productivity race are not the ones obsessively tracking token counts. They are the ones building cultures where AI fluency is expected, experimentation is encouraged, and outcomes (not inputs) determine performance.
What This Means for Growing Businesses
If you run a business with 20 to 200 employees, this trend will reach you within 12 to 18 months. Enterprise tools from Microsoft, Google, and Salesforce are all building AI usage analytics into their admin dashboards. Your team leads will soon have access to data showing who uses Copilot, who uses AI in Salesforce, and who ignores it entirely.
The question is whether you use that data wisely or let it become another vanity metric that drives the wrong behavior.
Three practical steps:
- Set outcome targets first. Before tracking AI usage, define what success looks like. Faster project delivery? Fewer support tickets? Higher proposal win rates? Make AI a means to those ends, not an end itself.
- Track adoption at the team level, not individual. This reduces surveillance anxiety while still identifying which teams need more training or support.
- Create safe spaces to experiment. Dedicate time for employees to try AI on real tasks without pressure. The best adoption happens when people discover value themselves, not when they are told a metric depends on it.
The Bottom Line
Tokens as a productivity metric is the "lines of code" debate of the AI era. It measures activity, not value. It is a useful signal for understanding adoption, but a dangerous KPI for evaluating performance.
Meta, Google, and the monitoring software industry are pushing measurement forward because the pressure to justify AI investments is real. But the smartest companies will measure what AI enables their people to achieve, not how many tokens they burn getting there.
The future of work is not about consuming more tokens. It is about creating more value with every token consumed.
