AI may save drafting time and still cost more in missed errors
Solomon Shalom Lijo’s preprint pairs time saved with the cost of errors that get through review. A reanalysis of developer data changes with the method used. In a small AI-reviewer test, a low-error-rate claim reduced checking and let wrong figures through. None of this proves a broad effect on human workplaces or schools.
What the source reports
Solomon Shalom Lijo’s preprint asks how much AI helps once missed errors count too. It proposes measuring time saved against errors that escape review and the cost of fixing them. A new look at public developer-study data gives different slowdown estimates depending on the calculation. In a small test with AI reviewers, a claim of low error rates cut checking by seven of nine figures. A wrong number reached 22.5 percent of briefs in that condition, versus none in the comparison. These results come from a model analysis, a reanalysis and a small AI test. They do not estimate effects in human workplaces or schools.
Original source
- Title
- Counting the Bandages: On measuring the time that AI saves
- Author
- Solomon Shalom Lijo
- Publication
- arXiv cs.AI
- Date
- Wednesday, October 28, 2026