Chapter 18: Cost Management

Agentic systems can be expensive. Every reasoning step, tool call, and generated token costs money. Without attention, costs can spiral.

Understanding the Cost Drivers

• Input tokens: Context sent to the model
• Output tokens: What the model generates
• Tool calls: External API costs plus processing
• Iterations: Complex reasoning multiplies all of the above

Cost Optimisation Strategies

1. Right-size your model: Not every task needs the most powerful model
2. Optimise context: Large contexts mean large costs. Keep context lean
3. Limit iterations: Set reasonable bounds on agent loops
4. Cache where possible: If the same queries recur, consider caching responses
5. Monitor and alert: Track cost per conversation. Set budgets and investigate anomalies

The Cost-Quality Trade-off

Cheaper isn't always better. A cheaper model that gives poor answers damages trust.