As AI takes on more work, organizations are understandably asking where agents can reason, act, transact, and move work forward with less human intervention.
There is another question we need to ask: where does human participation actually create value?
For years, people have served as connective tissue within enterprise processes: investigating exceptions, moving information between systems, chasing approvals, reconciling discrepancies, interpreting rules, coordinating handoffs, deciding what happens next, and checking whether something happened correctly.
AI increasingly has the potential to assume much of that work. That does not mean the goal should be to remove the human from the process. It means we need to understand why the human is there.
Not every human touchpoint is the same
Sometimes the human is there because technology historically could not carry the work forward. Someone has to gather information, move it, reconcile it, initiate the next step, or coordinate between systems and people. As AI becomes capable of reasoning and acting across systems, much of this work may no longer require human intervention.
Sometimes the human is compensating for a poorly designed process. The work is fragmented, rules are unclear, ownership is ambiguous, and exceptions have accumulated over time.
Automating those activities may make the process faster without making it better. We risk automating dysfunction.
What about when the human is creating value in the process? The person may be exercising judgment, accepting accountability, recognizing context, interpreting ambiguity, challenging an assumption, understanding another person, introducing productive friction, or making a consequential decision.
Human participation can also serve an important compliance, governance, policy, risk, or ethical purpose.
Compliance is not a rubber stamp
This distinction becomes particularly important in HR: hiring, compensation, promotion, performance, accommodations, employee relations, terminations, and workforce reductions.
AI may become increasingly capable of analyzing information, applying rules, identifying patterns, recommending actions, and executing portions of these processes. But capability alone does not answer whether a decision should be delegated.
Law, policy, governance, organizational risk, and ethics require controls, oversight, accountability, or meaningful human judgment. Simply inserting a human approval into an AI-enabled process does not necessarily accomplish that.
If a person merely rubber-stamps a recommendation AI has already made, little human value has been added.
Compliance cannot simply be bolted onto AI as a human approval step. It needs to be designed into the work itself: applicable rules, controls, decision rights, thresholds, documentation requirements, escalation conditions, and auditability.
When meaningful human judgment is required, we should be able to explain why. What judgment is the person expected to exercise? What information do they need? What responsibility are they assuming? What evidence of that decision should remain?
Otherwise, we may simply recreate an approval workflow with AI in front of it.
Mapping human value
Perhaps AI also requires us to rethink something organizations already know how to do: value-stream mapping.
Traditional value-stream mapping helps us understand how work flows, where value is created, where delays occur, where handoffs happen, and where waste exists.
What if we also mapped where human participation creates value?
As work moves through reasoning, decisions, execution, and outcomes, we could ask:
- What value must be created here?
- Can AI create that value reliably, appropriately, and compliantly?
- If human participation remains, what value is the human expected to create?
That last question may be particularly revealing. If we cannot explain the value of a human touchpoint, perhaps we should question why it remains.
But the inverse matters just as much. If we can articulate the human value required, whether judgment, accountability, empathy, interpretation, creativity, challenge, or contextual understanding, then the AI-enabled process should be intentionally designed to support that contribution rather than inadvertently diminish it.
As the work changes, the same analysis may reveal where human capability needs to grow or people need to upskill to participate effectively in redesigned work.
AI enablement should not only help us understand what work humans no longer need to perform. It should help us understand what humans may need to become better at.
Beyond human in the loop
We often describe responsible AI using the phrase human in the loop. But that tells us where the human sits. It does not tell us why the human is there.
Human in the loop describes placement.
Human value describes purpose.
Human oversight defines responsibility.
The future of work does not necessarily need humans in every loop. It needs humans intentionally designed into the places where human participation creates value. And where human participation does not create value, we should be equally willing to question why it remains.
The opportunity presented by AI is not simply to automate more work. It is to understand work more deeply, to distinguish the human activity we inherited from the human contribution we intentionally want to preserve.
Perhaps the more important question is not where humans remain in the process, but where being human matters to the value being created.
Lisa Insley
Founder | HR SOS
Frequently asked questions
What does meaningful human participation mean in an AI-enabled process?
Meaningful human participation means a person is there to create a specific kind of value, such as judgment, accountability, empathy, interpretation, challenge, or contextual understanding. It is not simply a required approval after a recommendation has already been made.
Why is a human approval step not always enough?
A person who only rubber-stamps an AI recommendation may add little value or accountability. Effective oversight needs clear decision rights, the information needed to exercise judgment, defined escalation conditions, and a record of the decision.
How can leaders decide where people should remain involved?
Leaders can map each stage of work and ask what value must be created, whether AI can create it reliably and appropriately, and what value a person is expected to add when human participation remains.