An optimization can become a loss when its engineering effort, verification, and rollout cost exceed the savings it is expected to produce. The break-even rule says to stop when further optimization costs more than it saves. A practical threshold is to avoid optimization projects when monthly AI tooling spend is below two weeks of one engineer’s fully loaded cost. section 9.2.8 section 9.1.3
Apply the rule by accounting for the complete change, not just its implementation:
- Compare expected recurring savings with engineering time and the cost of reviews, regression checks, and canary deployment. section 9.1.3
- Prefer zero-infrastructure behavior changes first: choose tools by task, narrow context with focused file references, and reduce unnecessary output. section 9.2.8 section 9.1.3
- Stop before pursuing the diminishing tail when a week of pipeline or tooling work saves only a small monthly amount. section 9.2.8
Choosing an optimization level
A workflow costing $8,754 per month can justify several improvements: roughly two weeks of changes reduce it to $1,407, an 84% reduction. By contrast, after the major savings, a week of CI pipeline engineering for $200 monthly savings fails the break-even test. For one pagination task, focused files, task-appropriate models, and diff-only output reduce cost from $1.44 to $0.074 while avoiding a retry. section 9.2.8
