Part 1 of 4 · 1 min read
Overview
An agent that works but loses money per task is not a product. Cost depends on steps, tokens, retries and failure handling, which are architecture decisions (Level 11), and revenue depends on how you meter value. This is the module that turns the engineering levers into a business case.
About the example, and what is real
Real: the cost-per-task, cost-per-resolved, price-floor, margin, customer-distribution, price calibration and sensitivity code, and the 17 tests in code/mini-pricing. Assumed: every input: $3 and $15 per million tokens, 28,350 input and 1,500 output tokens, 60 percent of input from cache at 10 percent of the price, $0.015 tools and infrastructure, resolution 55 percent, escalation 35 percent, failure 10 percent, $1.50 credit per failure, a $4 human task, the research workload, and the customer population (lognormal seats, usage and complexity). Not real: no market data, competitor prices or customer willingness to pay; do not use the dollar figures as price advice.