AI Agents by the Numbers: Step Success Rates, Retries, Tool-Call Costs, Human Hand-Offs & Cost per Completed Task
Plan an AI agent workflow before you build it: how step reliability compounds, what retries and tool calls cost, when to hand off to a human and what one completed task really costs, with every cell checked as you go.
One-time payment · lifetime access · lessons, labs, exam and certificate included · no subscription.
About this course
This course turns an AI agent workflow into numbers you can check before you build it. For a fictional invoice agent you multiply step success rates into end-to-end success and find the weakest step, calculate how retries lift success and what the expected extra attempts cost, and price each run from input tokens, output tokens and tool calls. You convert cost per attempt into cost per completed task for three designs, set a confidence threshold for human hand-offs and count the errors that slip through, check a latency budget with the 90th percentile, read step success rates from an evaluation log and finish with a business case, payback and chart.
Lessons
- Compounding Reliability: Step Success Rates & End-to-End Success Pro
- Retries: Success After k Attempts, Expected Attempts & Extra Cost Pro
- Tokens & Tool Calls: What One Attempt Costs Pro
- Cost per Completed Task: Paying for Failed Runs & Comparing Three Designs Pro
- Human Hand-Offs: Confidence Thresholds, Review Minutes & Slip-Through Errors Pro
- Latency Budgets: Retry Delays, Runs Over the Limit & the 90th Percentile Pro
- The Evaluation Log: Reached, Passed & Step Success Rates from 30 Test Tasks Pro
- The Business Case: Human vs Agent Cost per Month, Payback & the Chart Pro