How do you estimate a project when AI writes the code?

Short summary

AI hasn't killed software estimation—it's redistributed the uncertainty. The optimistic case gets dramatically faster with AI scaffolding, but the pessimistic case now must account for confidently wrong AI-generated code that hides bugs until integration. The most-likely case is hardest to pin down because developers lack calibrated experience with AI-assisted workflows. Three-point estimation (PERT) remains essential, but each anchor point needs rethinking.

• •AI lowers the optimistic estimate but makes pessimistic estimates harder to calibrate due to hidden integration failures
• •Three-point estimation (PERT) is still valid but all three numbers need recalibration for AI-assisted workflows
• •The real cost shift is from writing code to reviewing, debugging, and integrating AI-generated output

Generated with AI, which can make mistakes.