Principles / The Recovery Tax
§ Principle 21 of 23
The Recovery Tax
Failures in agentic systems have significant costs. Recovery from failures often exceeds the original task in time, tokens, and cognitive load.
The Recovery Tax means that prevention is dramatically cheaper than cure in agentic systems. When an agent fails — through State Loss, Execution Hallucination, or Plan Drift — the cost isn't just redoing the work. It includes: diagnosing what went wrong, understanding the gap between what was claimed and what actually happened, undoing any incorrect work, re-establishing the correct state, and then completing the original task with additional oversight. Each of these steps costs time and tokens, and the total typically exceeds the original task cost by 2-5x.
Why it matters
The Recovery Tax changes the ROI calculation for quality measures. Investing in Execution Integrity, Validation Gates, and Task Accountability isn't just about quality — it's about cost avoidance. Every failure prevented saves 2-5x its original task cost in recovery. This makes prevention investments some of the highest-ROI spending in any AI system.
In practice
Track your Recovery Cost per failure type. When you identify which failures are most expensive to recover from, invest in preventing those specific failures. The ROI is typically immediate and substantial.