In normal phased development, the cost to make a change (e.g.
fix a defect) rises exponentially as the phases progress.
This is a good argument for doing a lot of work to make sure
you leave the current phase with a product that's as close
to perfect as you can make it.
XP argues that you can use the advances made in software
technology to flatten the cost curve, and that this underwrites
many otherwise risky practices.
See ExponentialCostCurve, KentBecksChangeCostXpArticle, DecisionMathAndYagni, ChangePatternCategoryChange