Reproducibility is the load-bearing constraint in any empirical discipline. A finding that cannot be independently verified remains a hypothesis until it is. Economics is moving toward greater empiricism and invention as a field, but that shift is arriving alongside sharp questions about whether the work holds up to replication and about what role AI is playing in generating it.
The field's methodological direction carries a specific consequence. Theoretical work in economics has its own internal logic: assumptions and derivations are visible in the text. Empirical work is different. It depends on data and on the analytical choices made at every stage between collection and conclusion. When a study cannot be replicated, the failure can sit at any one of those points. A more inventive, more empirically active discipline means more of that surface area exists, and more results requiring independent confirmation.
AI adds a layer the field is still working through. When a model assists in shaping analysis or generating outputs, the chain of decisions a peer reviewer or independent replicator would need to audit becomes harder to trace. What the researcher contributed and what the model contributed is not always clear. The norms for disclosing and documenting AI's involvement in academic research are still being established, and economics is operating ahead of those norms.
The two concerns run together. A field producing more empirical work with AI assistance generates more replication targets and more questions about whether those targets are auditable at all. The watch is whether methodological standards for both can keep pace with the rate of output.