81% of researchers already use AI in their workflows. The question stopped being “if” about two years ago. The real question now is: can you document what the AI did?

Liao et al. (2024) surveyed 816 verified researchers. The adoption rate is 81%. That number is not inflated by tech optimists or early adopters — it reflects a broad shift in how empirical work actually gets done. Korinek (NBER WP34202, 2025) goes further and shows that LLM agents are already running full research workflows: literature scans, econometric code, robustness checks, draft synthesis.

The researcher+agent ensemble is the new unit of production in applied economics. That is not a prediction. It’s a description of current practice.

────

For private sector economists, consulting firms, and MDB teams, this has direct implications for how you structure, evaluate, and document research deliverables.

The capacity question has shifted. Asking whether a team “uses AI” is now as uninformative as asking whether they “use spreadsheets.” The meaningful questions are: which tasks are AI-augmented, how is that validated, and what does the human judgment layer look like?

The accountability gap is growing. When an agent runs a regression specification, selects control variables from a literature scan, and flags outliers — and the economist reviews the output — where does interpretive responsibility sit? This is not a philosophical question. It’s an audit question. And most current deliverable frameworks don’t address it.

Reproducibility has new requirements. A research memo produced with AI assistance that doesn’t document the prompts, model versions, and validation steps is, in a meaningful sense, less reproducible than one that does. Your quality standards should reflect that.

────

What I see working in practice with applied economics teams: treat the AI workflow as a methodology section, not a footnote. Document it with the same rigor you’d apply to data sources or estimation strategy.

The teams that build those habits now will be significantly more credible — and more auditable — when the standards catch up to the practice.

#AppliedEconomics #EvidenceBasedPolicy #AIResearch #InternationalDevelopment