Agent workflow guide
AI Agent Cost Calculator Guide
Agent costs can grow faster than a single chat request because each run may include planning, tool calls, retrieved context, intermediate reasoning, final answers, and retries.
Break the agent into billable steps
A useful estimate starts with the actual loop the agent follows. Count the first instruction prompt, each tool-planning step, each tool result sent back to the model, the final answer, and any validation or repair pass. Even when a user sees one answer, the model may have processed several request-response cycles.
Collect six usage numbers
- Base prompt tokens: system instructions, tool schemas, policies, examples, and stable product context.
- User task tokens: the user's request and any attached text for the run.
- Tool result tokens: search results, database rows, file excerpts, API responses, or code snippets returned to the model.
- Output tokens per step: planning notes, tool arguments, intermediate summaries, and the final response.
- Average steps per run: the number of model calls needed before the task finishes.
- Retry rate: extra calls caused by validation failures, timeouts, tool errors, or low-confidence answers.
The agent cost formula
Monthly planning formula
Input per run = first-call input + repeated step input + tool result tokens
Output per run = planning output + tool-call output + final answer output
Monthly cost = successful runs × (1 + retry rate) × model price for input and output tokens
Use the calculator for a baseline by entering the likely text and file volume, then multiply by the average number of model calls per task. Keep input and output separate because many providers price them differently.
Example estimate
Suppose a support agent starts with 8,000 base prompt tokens, receives a 600-token ticket, performs three tool calls, and sends 4,500 total tool-result tokens back through the model. If each run uses four model calls and generates 1,200 output tokens before the final answer is complete, one visible support answer can involve more than 30,000 processed input tokens before retries.
Reduce cost without weakening the agent
Shorten tool schemas, return only the fields the model needs, summarize long tool results before the next step, cap the number of search results, and route simple tasks to a lower-cost model. For stable instructions or tool definitions, compare the estimate with the prompt caching guide so repeated input is not over-counted.
Agents can also hit API capacity limits because one visible action may create several model calls. Use the token rate-limit guide to estimate peak requests per minute before launching high-volume workflows.
Measure the production loop
After launch, log model calls per user task, input tokens, output tokens, tool result size, retries, and whether the task succeeded. The most useful traffic and conversion improvements usually come from pages that help teams estimate these real operating costs before they choose an agent architecture.