https://aifaultline.com/authors/edward-mcfowland/New papers by Edward McFowland in the AI Fault Line guide, and news naming them. https://aifaultline.com/research/p/doi-10-1287-orsc-2025-21838/https://aifaultline.com/research/p/doi-10-1287-orsc-2025-21838/Sun, 01 Mar 2026 00:00:00 +0000A preregistered experiment with 758 Boston Consulting Group consultants finds that GPT-4 access improved performance on realistic tasks inside AI's capabilities but worsened it on a task just outside them. The authors call this uneven pattern a "jagged technological frontier" that can run through a single knowledge workflow. https://aifaultline.com/research/p/doi-10-2139-ssrn-4573321/https://aifaultline.com/research/p/doi-10-2139-ssrn-4573321/Sun, 01 Jan 2023 00:00:00 +0000In a field experiment with several hundred Boston Consulting Group consultants completing realistic case tasks, access to GPT-4 raised completion speed and quality substantially for tasks within the model's capability frontier, but consultants using AI on a task just outside that frontier were markedly more likely to produce wrong answers than those without it. The 'jagged frontier' finding — that AI's capability boundary is uneven and hard for users to perceive in the moment — has become a touchstone for warnings that AI productivity gains are highly task-dependent rather than universal.