Disney built an AI leaderboard. One employee called Claude 460,000 times in nine days.

Disney runs an internal leaderboard that ranks its product and tech staff by how much AI they use, and one engineer called Claude 460,600 times in nine days. Business Insider got the dashboard screenshots: 4,800 Disney Entertainment and ESPN employees, the most concrete picture yet of Fortune-100 AI consumption.

The headline number is not the only one that matters. Across the same nine mid-April workdays, the group logged about 1.4 million Claude invocations and burned 3.1 billion Claude tokens. Cursor, the agentic coding IDE Disney also pays for, drew even harder: 13.3 billion tokens. Business Insider’s analysis, corroborated by The Bridge Chronicle, pegs the implied enterprise spend at roughly $185,000 on Claude and $627,000 on Cursor for the window. Annualize that and the same team is on a path to north of $20 million on two AI vendors in a single year.

What the dashboard actually shows

The screenshots show a leaderboard, per-user request counts, token consumption, and “milestones” employees can unlock based on consecutive days of use. Streaks are the part that’s drawing attention internally. The pattern Business Insider’s reporting describes is best captured by an unnamed employee quoted in one of the secondary writeups: “They’re celebrating it now, but we’ll see how long that lasts.” Workers have started calling the behavior “tokenmaxxing”, a Slack-coined term for grinding the metrics that win the leaderboard.

The numbers also tell you what’s a real workflow and what’s a chatbot. 51,000 Claude calls a day, the top user’s pace, isn’t a person typing prompts. That’s an agent loop, almost certainly something like a long-running Cursor or Claude Code session firing many small tool calls per task. AI safety researcher Greg Sommer, interviewed by Business Insider, said the shape is consistent with agentic workloads. “It makes a lot of sense that you would see a couple of users having astronomically high consumption and requests relative to other users,” he told the outlet, the implication being that one engineer running parallel agents will dwarf a hundred peers using a chat sidebar.

The 234.2 million tokens that single user consumed averages roughly 510 tokens per request. That’s small for a chat session and exactly right for tool-call traffic where the model reads a snippet, decides on an action, and hands control back to the harness.

The cost shape this implies

For Disney, the per-employee cost on the engineering side is now visibly material. Spread the $812,000 nine-day combined Claude + Cursor estimate across the 4,800 staff and you get about $169 per person for that window. Cursor alone runs roughly $130 of that. Disney has its own paid SAML deals with both vendors, so retail card pricing isn’t the right number, but even at meaningful enterprise discounts the trend line is steep enough to explain why GitHub spent the same week pivoting Copilot to metered token billing and why Anthropic’s quality postmortem prompted a wholesale reset of usage limits for paying customers. Power users are the ones moving the bill.

It also clarifies why Cursor was in talks for a $50 billion valuation earlier this month. The Disney numbers say its enterprise gross dollars per developer per month run higher than Claude’s, and that gap mostly reflects how much tool-call traffic Cursor’s agent harness generates compared with Claude’s chat surface. Anthropic monetizes some of that anyway, since the underlying tokens are billed through its API, but Cursor is the one capturing the workflow lock-in.

What this means for you

If you’re running a similar AI rollout, three things in Disney’s setup are worth borrowing and one is worth avoiding. Borrow the per-user dashboard, because you cannot run a token budget you can’t see. Borrow the model-and-tool split (Claude for review and prose, Cursor or Claude Code for agentic edits), because the cost shape is so different that pooling them muddies the signal. Borrow the fact that they showed it to staff at all, because in a year where pricing changes weekly, opaque AI spend reads to engineers as either a coming layoff or a coming clawback.

What to skip is the leaderboard. Gamifying token use rewards the wrong thing. The metric that matters is whether the agent shipped working code or whether the prompt produced a usable artifact. A team that hits 51,000 calls a day might be the highest-output engineer in the company, or it might be the one burning 234 million tokens to chase a benchmark. Disney’s numbers don’t tell you which. Yours probably won’t either, unless the dashboard you build measures outcomes, not inputs.