Anthropic’s president weighs in on the tokenmaxxing debate

• Daniela Amodei waded into the discourse around companies overspending on AI.
• The Anthropic president is optimistic that much more capable AI models are on the way.
• Anthropic’s AI use is tracked without a leaderboard, she said.

Daniela Amodei has entered the tokenmaxxing discourse in Silicon Valley — and no, Anthropic does not have an AI leaderboard.

At Thursday’s Bloomberg Tech conference in San Francisco, the Anthropic president and co-founder took a question about tokenmaxxing, where developers use as much AI as possible and rack up huge bills with unclear business payoffs.

Amodei was diplomatic, but offered a bullish vision.

She said that though AI models have dramatically improved over the last two years, this isn’t the end state of their progress.

“I actually think there’s a lot more distance to go still for what the models will be able to do two to four, six to eight years in the future,” Amodei said.

The executive also said businesses will find new ways to use the tools, and learn together as a larger community.

She acknowledged that workers are reckoning with a new paradigm, where they’re directed to use AI and sometimes even tracked by their employers.

“Today there’s this feeling that’s like, ‘Oh, like AI, you know, the leaderboards, and it’s like I have to use it, and what am I going to use it for?'” Amodei said. “My hope is that over time it’ll be more incorporated into the day-to-day of how humans do our work, how we communicate together, and that there will actually be a lot more value realized in a way that feels really good to people.”

Anthropic does not have a token-usage leaderboard, she noted, though the company does track general use of the company’s Claude products as teams work.

“But there’s not like, ‘You must use AI and you must use Claude’,” she added.

Claude Code, Anthropic’s juggernaut AI-coding product, is one of the main drivers of the tokenmaxxing discourse.

The tool uses significantly more tokens — the units of AI input and output — than a classic chatbot interface. That gap grew more substantial with the advent of agents, which a user can direct to autonomously churn through tasks. Anthropic and OpenAI partly charge on a per-token basis, so the boom in AI coding has fueled a surge in revenue.

Companies are navigating how to spend on AI, after some executives at top companies encouraged a freewheeling attitude. Amazon recently shut down an informal, employee-made leaderboard that tracked the use of AI tokens.

“Please don’t use AI just for the sake of using AI,” Dave Treadwell, an Amazon senior vice president, told staff.

A similar leaderboard at Meta was taken down by the employee who made it, The Information reported. It had been titled “Claudeonomics.”

Have a tip? Contact this reporter via email at scouncil@insider.com, or over text, Signal, Telegram, or WhatsApp at 415-757-8198. Use a personal email address, a nonwork WiFi network, and a nonwork device; here’s our guide to sharing information securely.