Habr explains why token price doesn't explain AI agent economics in business

Habr published an article analyzing the economics of AI agents in corporate customer service — a topic that, in the author's opinion, usually boils down to a simplified discussion of token pricing.

What questions are typically asked about agent economics

The author lists the typical set of questions that arise when discussing costs of AI agents: how much does a million input and output tokens cost, what will the budget spending rate be, which model to choose to avoid overpaying, and how many GPU accelerators will be needed when deployed in the customer's infrastructure — along with their utilization.

• Discussion of agent economics usually focuses on token pricing
• Budget spending rate and model choice are considered
• When deploying in customer infrastructure, the number of GPUs and their utilization are calculated
• The author argues that this is insufficient for corporate customer service

Why token pricing alone is not enough

The author emphasizes that all the listed questions are correct, but when it comes specifically to corporate customer service, the answers to them are critically insufficient for a complete picture of economics. A full assessment requires accounting for additional factors beyond direct computation costs — including integrations, people, and the learning curve of the system itself, which the article's title alludes to.

What this means

The article reminds us that when implementing AI agents in business processes, focusing solely on token pricing is a mistake: real economics consist of integration costs, changing roles of people in the process, and time spent for the system to reach operational efficiency.