Token Starvation Fallout

This token starvation has put me on a path to understand the broader context of what this can mean.

Not just for me or any given worker using AI agents to do stuff, but for industry as a whole. Yes, this happened to me, but its not a blip. I’m not the only one.

I’ve seen stats indicating that heavy users of agent-based software are spending several thousand dollars per day on tokens.

Here’s a real-world example: today (I’m not token-starved today), I’ve used 207 million tokens. 207 million. The cost estimate for that is ~$1600. Just me. Just today. I cost my company $1600 in tokens above my loaded personnel costs.

That’s a hard pill to swallow.

So, what can we extrapolate, if anything, from worker token starvation?

1. This will limit and slow down the existential risk to human knowledge workers’ jobs. Until compute gets cheaper and more abundant.
2. CFO’s will have to figure out token budgets, including whether they should hard limit or simply predict.
3. Leadership will have to develop a rubric for assessing worker performance, which will include token usage. Does more indicate more productivity or more waste?
4. Is token budget part of what a worker negotiates when deciding whether to take a job?
5. Leadership will have to manage token budget allocations.
6. Workers will have to manage their own token budgets and find ways to use them efficiently.

The world is a-changin’.