AI Shifts Software Estimation Uncertainty, Not Eliminates It

AI-assisted coding improves optimistic estimates but widens pessimistic ones, as generated code often fails at system seams. Three-point estimation remains useful, but developers must resist quoting the seductive best-case number. The technique's value grows as AI redistributes work into less predictable failure modes.