Essay argues token costs are now too cheap to meter

A widely discussed essay makes a simple argument. AI inference has become cheap enough that charging customers by tokens, prompts, or individual AI actions may stop making commercial sense.

The post has drawn attention on Hacker News, with 167 points and 134 comments, because it challenges a pricing model many AI products still use. If the underlying cost of serving a request keeps falling, usage billing can become more about managing customer behaviour than recovering a meaningful variable cost. That creates room for flat monthly pricing, retainers, or plans based on outcomes and service levels instead.

For operators, this is a useful question to ask of any AI tool or service you sell or buy. Is usage genuinely the cost driver, or is it just the easiest thing to count? A token-based bill can make a product feel transparent, but it can also create uncertainty for customers who want to know their monthly spend. On the other hand, unlimited plans need sensible boundaries, because even cheap usage is not the same as zero cost.

The likely winners are businesses that can price around real customer value while keeping their own delivery costs predictable. This is the kind of thing we build into an AI command centre, where the focus is on managing useful business workflows, not watching token counters.