An HR Director Built 8 AI Agents in One Week. Here’s What She Learned Not to Automate.

An Inc.com profile examines one HR director's rapid experimentation with agentic AI, building eight AI agents in a single week to handle different HR workflows. The story is notable less for what the agents did well than for what the experiment revealed about where AI agents should not be deployed in human-facing roles.

According to McKinsey data cited in the piece, HR professionals spend roughly 57% of their time on administrative tasks, making the function an obvious target for agentic automation. But the director's experiment surfaced a distinction that is likely to become central to responsible AI deployment in people-facing roles: some tasks are administrative in nature but deeply personal in context.

Key Takeaways

• Administrative volume is real: 57% of HR time on administrative tasks (McKinsey) creates a genuine automation target, and agent deployment for scheduling, benefits Q&A, and onboarding documentation showed clear efficiency gains.
• Sensitive conversations can’t be automated: The director found that agents handling performance conversations, accommodation requests, and termination-adjacent communications produced outcomes that felt dehumanizing to employees, regardless of accuracy.
• Speed vs. trust tradeoff: Agents that responded faster than a human could were experienced by employees as "not caring", suggesting response latency is a trust signal in people-facing interactions, not just a performance metric.
• High-judgment tasks require human presence: Agents could surface relevant policy and options for complex HR situations, but employees needed a human to be accountable for the decision, not just the information.

The experiment provides a practitioner-level data point on where the agentic AI deployment boundary should sit in HR: automate the administrative container, not the human interaction inside it.

Read the full article on Inc.com