Editor's Note
Amazon built a leaderboard to track how many AI tokens its developers consumed. Employees gamed it, a practice Fortune named "tokenmaxxing" in May. One day after Fortune declared the experiment dead on May 28, Amazon pulled the leaderboard entirely. The cycle from mandate to collapse took less than a month, but the measurement problem it revealed has been building for years: across Fortune 500 boardrooms, enterprise AI programs are being evaluated against metrics that were never designed to capture actual value.
When Amazon tied developer performance to token consumption, it didn't measure productivity, it measured compliance, and its workforce responded accordingly.
Why It Matters: Tokenmaxxing is what happens when you deploy AI without redesigning the workflows and incentive structures around it, and employees who move faster with no reason to take on more work produce the same output at higher cost.
The problem with tokenmaxxing isn't just that employees gamed the metric. It's that the metric was a proxy for something nobody had clearly defined. That ambiguity runs deeper than any single leaderboard.
A survey of 349 technical workers found median self-reported productivity gains of 1.4 to 2x, but METR's own controlled research found that people overestimated AI's effect on their work by 40 percentage points on average.
Why It Matters: Self-reported productivity data is structurally unreliable, respondents have every incentive to look good, "productivity" means something different at every company, and enterprise deployments are too varied for survey-level generalization to hold without clearly defined value metrics established upfront.
Speed, it turns out, is the wrong variable. What actually matters is capacity: whether an organization can do things it couldn't do before, or sustain the same output with fewer resources. Almost nobody is measuring that. At the executive level, the gap between what's being tracked and what actually matters is even wider.
2026 AI Impact Survey Report: Grant Thornton
Grant Thornton's survey of Fortune 500 executives finds that two-thirds are measuring AI ROI through estimates rather than financial results, and 78% couldn't pass a basic AI governance audit.
Why It Matters: The measurement failure isn't confined to middle management, it reaches the boardroom, and without proper knowledge management and information governance infrastructure in place, AI systems will produce unreliable outputs regardless of how much is spent on the underlying models.
If the governance gap explains why measurement fails, the Deloitte data shows what the cost looks like in revenue terms.
Deloitte's 2026 State of AI in the Enterprise report finds that while 74% of organizations expect AI to grow revenue, only 20% are actually seeing that outcome today, because most are tracking productivity proxies rather than financial results and treating governance as an afterthought rather than infrastructure.
Why It Matters: The gap between individual productivity claims and organizational revenue impact is the difference between a vanity metric and a business result, and the companies closing that gap invest in reskilling and knowledge infrastructure before they scale, not after.
All of this assumes employees are actually using the tools in the first place. That assumption is increasingly fragile.
Writer's 2026 enterprise AI adoption report finds that despite heavy investment, 79% of organizations face significant adoption challenges, in large part because the conversation has focused on customer outcomes while consistently underweighting whether employees trust the tools they've been given.
Why It Matters: Trust is the adoption variable that never appears on a dashboard, but if employees don't believe in the tools, low-quality usage produces low-quality outcomes regardless of the underlying model, and that is the same dynamic that produced tokenmaxxing in the first place.
The pattern across all five of this week's readings is the same: organizations deployed AI and then tried to figure out what success looked like. Token leaderboards, self-reported surveys, executive estimates, productivity proxies, all of them are attempts to impose measurement on programs that were never designed around outcomes. For founders building now, the structural advantage is that you haven't made those commitments yet.
Define what you're trying to change before you pick a tool. Define how you'll know it worked before you deploy. The companies winning on AI right now aren't the ones with the most tokens or the most pilots. They're the ones that asked the right questions before they started counting.
If this issue resonated, send it to whoever in your organization is about to build an AI usage dashboard. Preferably before they start.
Have a take on any of this? Reply directly, I read every response.
Until next Tuesday,
Charles Costa, MLIS