You Added AI. The Workflow Still Needs Work.

This Lab Notes issue revisits stories from AI for HR News #002, originally published April 1, 2026.

Put an AI assistant into a slow approval process and you may get a draft in seconds. It can still sit in someone’s inbox for a week.

That is the thread running through this issue: AI capability is moving quickly, while the decisions around it still need human attention. Who sets the rules? How do we check whether an agent chose the right tool? What counts as evidence that the whole workflow improved?

News worth tracking

HR earns influence through the work

HR Executive argues that HR must lead in the AI era.

Our take: We would start with the decisions HR should help make: what employee data can enter these systems, how people can challenge an AI-assisted decision, and what managers need to learn before using the tools. Owning that work gives HR a concrete role in AI strategy.

Ask how the agent chooses its tools

Cheney Zhang makes the case for skills as an agent tool interface, raising questions about how agents find and use tools as their options grow.

Our take: A demo with three tools tells you little about an agent facing dozens of overlapping options. We would ask vendors to show how they test tool selection, permission boundaries, and recovery when the agent gets it wrong, using tasks that resemble your HR workflow.

Watch what people do with the AI

ResearchChatAI, introduced by Roman Briker and colleagues in The Leadership Quarterly, is an open-source platform for experiments with people interacting with LLM agents.

Our take: A survey about willingness to use AI and a record of what someone actually does with it answer different questions. For a pilot, we would look at when people check an answer, correct it, or accept it without review; those behaviors help show where the workflow needs better support.

Plan for the capabilities the workflow needs

Sangeet Paul Choudary’s Kyndryl Institute essay argues for thinking about agents as capabilities rather than digital workers.

Our take: Calling an agent a digital employee makes it tempting to drop it into an existing org chart. We would map what the workflow needs to accomplish, then decide where AI fits and where human judgment, relationships, and exception handling become more valuable.

Measure what the saved time makes possible

The AI Daily Brief episode on AI’s “second moment” discusses changes in deployment and reported value.

Our take: “We saved two hours” is a useful starting point. We would also ask what happened next: did a recruiter spend more time with candidates, did a manager get better support, or did the time disappear into more administrative work?

One useful move

Pick one workflow where the team already uses AI. Draw the steps from request to finished result, including approvals, checks, and handoffs.

Mark where AI saves time and where work still waits. Choose one measure for the full workflow, such as turnaround time or corrections needed, and one improvement to try. A faster first draft is useful; this exercise tells you whether the benefit reaches the person waiting for the result.

A lighter thought

Before you give your AI agent a job title, give it an escalation path. “Ask a human” needs to name an actual human.

Question for the team

Where are you still inserting AI into an existing workflow instead of redesigning the system around it? What would it take to change that?