What it means in practice
A token is the piece of text a model works with. Token consumption is therefore how much goes through the model, and the invoice follows from it.
Concrete numbers from practice: one sales company we work with currently spends €1,000 to €2,000 a month per software engineer. At first glance that looks like a lot. At second glance it is a fraction of what that same person costs.
Why low consumption is not good news
Management instinctively reads consumption as a cost to be squeezed. In practice it behaves more like an indicator of learning.
If a team burns through its weekly limit, it usually means they are genuinely working with AI. If consumption is flat, either people have nothing to work on or, more likely, they do not know what is possible.
How to think about it
Count it as an investment in a tool, not an operating line. The useful question is not “how much did we spend?” but “what came out of it?” Twenty website variants for a thousand dollars in tokens is a different conversation from an agency tender.
The cost side of agentic engineering is one of the topics we work through in the workshops with management and developers alike, because each needs different conclusions from the same number.