Agent workflow cost
Model the cost of a multi-step AI agent
For teams designing agents with conditional classification, retrieval, tool, and generation steps.
Open this preset in the workbenchDecision
What this route helps you decide
Find the submitted and successful-task cost of a probability-weighted workflow rather than treating an agent as one call.
Inputs
- Execution probability for every workflow step
- Executions and tokens consumed per step
- Step-level cache, failure, and retry behavior
- Synthetic or user-entered model rates in USD
Outputs
- Probability-weighted submitted workflow cost
- Retry-adjusted successful-task cost
- Typical active-user cost per month
- Modeled margin for the complete AI workflow
Worked example
Worked example
A three-step support agent classifies every task, retrieves context conditionally, and generates a final answer with bounded retries.
- Cost / successful task
- $0.0387
- Typical user / month
- $4.6399
- Modeled margin
- 63.1%
Assumptions
- Each step has an explicit execution probability
- Blocking success compounds across steps
- All rates are synthetic
What this does not decide
- Does not measure output quality
- Does not simulate tool latency
- Does not select a model
Reproduce the method
All values use synthetic or user-entered rates and deterministic formula version 2.0.0. Review the calculation methodology and source policy.