Disney wants its streaming engineers to move faster with AI, but the company is already learning that raw token use is a poor substitute for better software.
Disney's latest AI push is not about a flashy product demo or a new chatbot with a name. It is about engineers, code reviews, token meters, and the less glamorous question now facing large companies: how do you make AI useful without turning usage itself into the job?
Business Insider reported on June 13 that Disney's streaming technology leaders have been pressing employees to use AI coding tools more aggressively, while warning against \"tokenmaxxing,\" the habit of burning through AI tokens simply to show activity. Andre Rohe, Disney's executive vice president of product engineering, raised the issue in a Wednesday meeting, according to the report. Staffers described the message as having three parts: increase the speed of feature and code delivery, watch inefficient token use, and keep code quality and resiliency intact.
That combination is the real story. Disney is not backing away from AI. It is trying to make AI behave like an operating system for engineering work, with budgets, dashboards, managers, and consequences. The novelty is not that engineers are using Claude or Cursor. The novelty is that a media company with one of the most valuable consumer brands in the world is treating AI coding as something to be managed in the same way it would manage cloud spend or production reliability.
Disney already gives some employees access to AI tools including Claude and Cursor, and Business Insider previously reported that the company has an internal AI Adoption Dashboard showing active users, requests, and token consumption. In April, the same outlet reported that one Disney employee invoked Claude about 460,000 times over nine workdays, roughly 51,000 times a day. Staffers said that level of use likely came from autonomous agents rather than a person manually typing prompts all day.
That is a useful number, but it is also a trap. A dashboard can tell a manager who is using tokens. It cannot, on its own, tell whether the resulting code is cleaner, safer, easier to maintain, or worth shipping. If a team learns that management is watching token consumption, some people will naturally push the number up. Engineers are very good at optimizing systems. They are also very good at optimizing the wrong metric when the metric is made too visible.
Disney appears aware of that problem. Business Insider reported that a person familiar with the company's strategy said the dashboard was meant to help employees use AI tools efficiently and effectively, not to reward high usage for its own sake. That distinction matters more than it sounds. In software, bad incentives do not stay theoretical for long. They show up as bloated pull requests, fragile generated tests, duplicated logic, and code that looks productive in a weekly report but becomes expensive in production.
The new AI budget fight is inside the workflow
The same debate is spreading beyond Disney. Business Insider reported that Paramount Skydance told technology staffers on Wednesday it would introduce per-user monthly spend limits on AI tokens, although an executive described the cap as high. The company has also been showing employees examples of AI cutting technical work from hours or days to minutes. That is the tension every large engineering organization is now trying to hold at once. Use more AI where it saves real time. Stop paying for noise.
Microsoft has been pulled into the same conversation, with Business Insider noting that Satya Nadella recently described tokenmaxxing as addictive. Replit's AI chief Michele Catasta called token leaderboards \"very dystopian\" in another Business Insider report this week. Legora's chief technology officer Jacob Lauritzen called tokenmaxxing a poor way to encourage AI use, arguing that companies should reward visible productivity gains rather than raw consumption. The names differ, but the complaint is the same: tokens are an input, not an outcome.
For Disney, the stakes are sharper because streaming software sits directly under the customer experience. A bad internal dashboard might waste money. Bad generated code in a streaming product can hurt playback, recommendations, account flows, advertising systems, or the reliability of Disney+, Hulu, and ESPN-related technology. A media company can tolerate experimentation in a sandbox. It cannot let a token race leak into systems that millions of subscribers touch.
This also lands at an awkward moment for Disney's broader AI strategy. Business Insider reported in March that Disney's planned OpenAI partnership fell apart after OpenAI moved to discontinue Sora, the video product that would have been tied to Disney characters and short-form content on Disney+. The company is still expected to use AI, but the center of gravity has shifted. Instead of one big external partnership carrying the story, the work is moving inside the company, where engineers are being asked to use AI every day and prove it is making the machine run better.
That is probably where enterprise AI was always heading. The pilot phase rewarded enthusiasm. The next phase rewards judgment. A manager who only asks why an engineer is not using AI enough will create one kind of behavior. A manager who asks what shipped faster, what broke less often, and what token spend bought the team will create another.
Disney's warning against tokenmaxxing is therefore not a rejection of AI coding tools. It is a sign that AI adoption has become normal enough to need discipline. The dashboard is no longer just a sign of ambition. It is also a control panel, and the hard part is deciding which lights on it actually mean progress.
Also read: UAL's Nerve Lab puts children's screen time under a sharper lens • AI agents are becoming guards for EV charging networks • China puts financial data closer to the center of state control
