Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of Artificial Intelligence on Knowledge Worker Productivity and Quality

Summary

A preregistered field experiment with Boston Consulting Group that introduces and studies the concept of a "jagged technology frontier" — the observation that AI assistance can improve performance on some tasks while worsening it on others of seemingly similar difficulty, even within the same knowledge workflow. The authors developed realistic management-consulting tasks and examined how knowledge workers' performance changes when they use AI to perform complex, knowledge-intensive work.

Contribution

Names and operationalises the jagged frontier concept and provides field-experimental evidence on its existence and managerial implications: AI use can shift both productivity and quality, but in direction- and task-specific ways that resist a uniform "AI makes work better" generalisation.

Method

Preregistered field experiment in collaboration with Boston Consulting Group using realistic consulting tasks; treatment-control comparison of human performance with and without AI assistance.

Relevance to RISE

The "jagged frontier" framing is directly applicable to evaluating RISE pipelines: it warns that aggregate productivity gains from agentic-research systems may hide systematic per-task degradation, exactly the pattern RISE evaluation hubs (e.g. asta-bench) need to detect. Methodologically, it is a template for the kind of preregistered field study that the RISE programme aspires to apply to autonomous research agents.

Critique / open questions

Tasks are management-consulting exercises designed for the study, not in-the-wild research workflows; the BCG collaboration provides generalisable knowledge-work conditions but limits direct transfer to scientific research. As with any single field experiment, replication across organisations and AI generations is open.

Key quotes

"We introduce and study the concept of a 'jagged technology frontier' to describe the uneven impact of artificial intelligence (AI) capabilities, where AI assistance improves performance for some tasks but worsens it for others, even within the same knowledge workflow and with a seemingly similar level of difficulty."