Researchers can deploy autonomous AI agents to execute a fixed coding instrument across countries/years, find and cite primary sources, and assemble structured datasets with documented gaps and uncertainty. A two‑stage design (instrument design + agent implementation) plus evidence policies lets small teams (or a single subscription) produce research‑grade data at a fraction of current cost.
— If adopted widely, agentized dataset construction could lower barriers to empirical research, shift demand for human research assistants, and change which claims get defended by traceable evidence.
Tyler Cowen
2026.05.12
100% relevant
The NBER working paper (Afonso et al.)’s DRIL run produced 129 sources and 136 evidence records for a 2025 tax‑expenditure update across eight countries at the cost comparable to a few hours of RA work.
← Back to all ideas