Reid Hoffman is right to suggest that AI token usage can serve as a useful adoption signal, but he's equally right to stop short of calling it a productivity metric. That distinction matters, because the gap between "people are using the tool" and "people are getting more done" is where most digital initiatives go to die. We'd go further: treating token counts as a proxy for value is not just imprecise, it's actively misleading for teams trying to make smart decisions about their AI investments.
Here's what that means for you in practical terms. If you're a team lead or a finance leader reviewing a dashboard that shows token consumption climbing month over month, it's tempting to read that as a win. But token volume only tells you that something is being processed, not whether that processing is moving the needle on your actual work. A marketing team might burn through thousands of tokens generating drafts that still require heavy editing. An analyst might use AI to produce a dozen variations of a report when one solid version would have sufficed. In those cases, higher token usage could actually signal inefficiency, not progress. Hoffman's caution is a reminder to pair usage data with context: what problem was the tool asked to solve, and did solving it change the outcome?
The practical takeaway is to treat token insights as one input among several, not the headline number in your weekly review. When you're evaluating whether AI is working for your organization, look for evidence of time saved on specific tasks, quality improvements in the final output, or the ability to take on work that was previously impossible within your resource constraints. Those are the metrics that connect directly to productivity, and they require qualitative follow-up, not just quantitative tracking. Ask your team what they're using AI for, why they chose that tool for that task, and what they would have done otherwise. That's where the signal gets useful.
So by all means, monitor token usage as a leading indicator of adoption. Just don't confuse activity with achievement. The teams that get this right will be the ones who build a feedback loop between usage data and real-world outcomes, adjusting their workflows based on what the tokens actually help them accomplish. That's not a flashy conclusion, but it's the one that will save you from investing in a tool that's merely busy, not beneficial. Start with the questions, not the charts, and you'll have a far clearer picture of whether your AI journey is heading somewhere worth going.