Daniel Litt argues that current AI limitations are temporary, expecting AI to autonomously resolve open conjectures as early as late 2026, though human mathematicians will remain essential.