How To Divide Work Between Humans And AI Agents

Manu Khetan is Founder and CEO of Rolling Arrays, a LinkedIn ‘Top Voice’ Influencer and Creator of the R7 Framework.

How do I divide work between humans and agents? Every leader I meet is asking some version of that question. Usually it arrives dressed up as a technology question, which is why it goes wrong.

In Part 1 of this series, I wrote about the blind spot: On a spreadsheet, value and overhead are the same number. This piece is about the fix. How you actually draw the line between what agents can take and what stays human. I have spent 17 years implementing HR technology. My own 400-person, six-country firm is adopting AI right now.​

The Unit Is The Task, Not The Job

The first mistake is treating the job as the unit. A job is a bundle of tasks, and the tasks inside one job vary wildly. Some are routine. Some carry judgment. One or two carry the trust the whole role exists for.

So you cut roles by task, not people by salary. That single sentence is most of the method.

We did this to ourselves first. Our delivery organization has 18 roles, and we read them task by task instead of by headcount. Two people carrying the same title turned out to be doing quite different work. The title told us very little. The task mix told us most of what we needed.

Two Questions Per Task

In the R7 Framework, we read every task on two axes.

The first question is about the nature of the work. Is this task physical, cognitive or relational? Doing work, thinking work or work that runs through a relationship.

The second question is about ownership. Is this task execution, judgment or trust? Execution is the routine doing, gathering, drafting, reconciling, monitoring. Judgment is deciding what is worth doing when the playbook does not fit. Trust is owning the outcome and holding the relationship, putting your name on it.

The Gate

Here is where the discipline comes in and where most programs skip a step.