A short blog post estimates the monthly cost of a state-of-the-art AI email agent at $22–$130, with a middle case of $26/month.

A software company targeting 75% gross margin would need to charge roughly $350/year before hosting and serving costs, leading to a list price around $500/year with a 15% volume discount.

By comparison, a Google Enterprise plan runs $11–$18/month, so a fully agentic solution would cost about twice that.

Smaller models can cut cost by 10–20×, and running models locally on the user’s GPU drops the marginal cost to zero.

The post argues that cost optimization will define the next 12–24 months of AI software: identifying which components can be executed deterministically (e.g.

, email filters as simple rules) and matching model size to workload.

With basic heuristics the overall cost can be reduced by 100×, and given GPU shortages this segmentation of inference is inevitable.