A recent Stanford study surveyed more than 1,500 professionals across 104 fields, asking them directly which tasks they would like AI agents to automate. The findings are striking: nearly half of Y Combinator-backed startups are focusing on automating activities that employees actually don’t want to hand over.
Rather than being replaced, people want to be freed from administrative burdens (scheduling, data entry, repetitive clicks) while staying deeply involved where creativity, judgment, and meaning matter.
To gauge how much control humans should retain over each type of activity, the researchers introduced the Human Agency Scale (HAS), a five-level spectrum:
- H1 – AI acts completely autonomously
- H2 – AI needs only minimal intervention
- H3 – Human and AI collaborate as equals
- H4 – AI largely depends on human input
- H5 – AI cannot function without the human
Levels of Human Agency Scale (HAS). Source.
For companies building AI agents, the takeaway is clear: trust is the decisive factor. If AI products ignore users’ real preferences and chase only the spectacle of a demo, they risk delivering solutions no one will adopt.
Read the full study here: https://arxiv.org/html/2506.06576v2.
Photo source Unsplash.