Tracking AI correction time: 78% of fixes are preventable project-mismatch issues

Short summary

A developer tracked time spent correcting AI-generated code across ~30 sessions and found 78% of correction time went to project-specific mismatches that explicit rules could have prevented — not logic errors. Only 8% was genuine AI mistakes. The invisible cost of corrections doesn't appear in sprint boards or velocity metrics, leading teams to underestimate the real cost of AI without codified project rules. The finding suggests investing in rule definition yields outsized returns.

• •78% of AI correction time was project-mismatch issues preventable by rules, not logic errors
• •Correction cost is invisible — it doesn't show up in sprint boards or velocity metrics
• •Defining project-specific conventions for AI could recover weeks of annual productive time

Generated with AI, which can make mistakes.