Apparently, Everyone Is Burning Through Tokens

In The Army Is Burning Through Its AI Tokens, Vittoria Elliott reports that the Army exhausted an annual pool of AI tokens shortly after encouraging widespread use of the technology.

“Apparently the whole Army burned through the whole year of tokens for just one service.”

Employees were initially allocated at least 200,000 tokens each month and automatically received more when they ran out. By mid-June, the shared pool was exhausted and usage limits had to be restored.

I am glad I am not the only one burning through tokens faster than expected. I do everything I can to reduce context, divide work into smaller tasks, and avoid making agents repeatedly process information they do not need. The tokens still disappear remarkably quickly.

Part of the problem is that useful agentic work naturally involves reading files, reviewing earlier decisions, running tests, inspecting failures, and trying again. That is where these tools become valuable, but it is also where their consumption becomes difficult to predict or control.

AI companies keep encouraging us to use these tools for more of our work while imposing limits that make sustained use increasingly frustrating. Apparently, even the United States Army cannot make the token budget work.

Read Vittoria Elliott’s full report, The Army Is Burning Through Its AI Tokens.