AI agents can run fieldwork chores. Judgment is still yours

From copilots to agents

Last year's AI helped you write. This year's agents do multi-step jobs: draft a questionnaire from a brief, watch live fieldwork, flag quality problems, rough out a first report. Supervised well, they take days of production work down to hours.

Jobs agents handle well right now

• Survey programming: Brief in, routed questionnaire with quotas out — checked by a human before launch.
• Fieldwork watch: Spotting incidence drops and rebalancing sample across sources mid-field.
• Open-end coding: First-pass classification against your codeframe; people take the edge cases.
• First-draft reporting: Chart-ready summaries an analyst turns into something a client can use.

The human job didn't shrink — it moved

Agents follow the brief, the guardrails, and the quality bar you set. Vague briefs now fail faster and at scale. The teams getting value treat agents like sharp juniors: give precise instructions, check the work, and keep methodology and storytelling with senior people.

If you're buying research this year

Ask suppliers three things: how their agents are validated, how automated output is labeled, and where a human reviews before anything reaches you. Anyone who can't answer clearly isn't ready to use agents on your budget.