The New Skill: Knowing What to Delegate to AI 

The hardest part of using AI at work may not be figuring out what it can do. It may be deciding what you are comfortable letting it do.

Let us think about how delegation normally works. A manager might ask someone to prepare a report but still want to look at it before it goes out. An HR team may automate a routine employee request, but want a person involved when the situation is sensitive. A finance team might let a system flag something unusual, while keeping the final decision with a human.

We have been making these calls with people for years. We just call it delegation.

AI has added a new choice to the mix. Now we must decide not only who should do something, but whether a person needs to do it at all, how much context AI needs, what it should be allowed to act on and when someone needs to step back in.

That is a different skill from simply knowing how to use AI.

A good manager does not try to keep every ball in the air themselves. They know which tasks need their attention, and which ones can be handed to someone else.

The same thinking is starting to apply to AI.

A report might not need to be prepared manually anymore. An email could be drafted without someone staring at a blank screen for twenty minutes. A straightforward employee query could be answered automatically, while anything unusual gets passed to a person.

On the surface, these sound like simple automation decisions.

They are not always.

The real question is what happens after the task leaves your hands.

If AI is preparing a report, who checks whether it has understood the numbers correctly? If it is responding to an employee, does it understand the situation well enough? If it is taking an action rather than just making a recommendation, what happens when something falls outside the usual process?

This is where judgement starts to matter.

It is tempting to look at a thorough list of repetitive tasks and think, “Let’s automate all of this.”

Fair enough. It is often a good place to start.

But repetitive does not always mean low-risk, and complicated does not always mean unsuitable for AI.

In the case of HR. An AI system could help sort employee requests, find information in policies, prepare summaries, or find something that needs attention. There is a lot of useful work it can take off someone’s plate.

But suppose an employee raises a sensitive concern about their manager.

AI may be perfectly capable of reading the information and organising it. That does not mean it should decide what happens next.

The difference is subtle, but important.

The point isn’t to hand over as much work as possible. It is to know which work is worth handing over.

This is where I think the conversation around AI can sometimes become too black and white.

We talk about AI doing a task as though the only choices are “AI does it” or “a person does it.”

Work rarely works that neatly.

You might ask AI to gather the information, while you make the decision. It might prepare a recommendation that someone reviews. It might handle a routine process on its own but stop and ask for approval when something doesn’t look right.

Consider something as ordinary as onboarding a new employee.

There are plenty of steps that do not need someone chasing them manually. Information can be collected, reminders can be sent, tasks can be triggered and outstanding items can be tracked.

But if something doesn’t fit the standard process, a person may still need to step in.

That isn’t AI replacing the person. It is simply deciding how much of the process AI can own.

And that is likely to become a much more major decision as AI agents move from answering questions to taking action.

There has been a lot of talk about becoming good at prompting AI. And yes, knowing how to communicate with these systems helps.

But there is another skill underneath it that may matter even more: knowing when not to use AI.

You need to understand your own work well enough to recognise what can safely be handed over, what needs more context and what still requires human judgement.

That means looking closely at the work people do every day.

Which tasks eat up time for no particularly good reason? Which decisions genuinely need experience? Where do mistakes usually happen? Which steps exist because they are useful, and which ones are simply there because “that’s how we’ve always done it”?

AI might make those questions harder to ignore.

And that’s a good thing.

Once you start looking at AI this way, it stops being just another tool employees must learn.

It becomes part of how work itself is organised.

For HR, that matters quite a bit. Employee information, attendance, leave, payroll, timesheets, and other processes don’t really exist in isolation. They overlap, depend on one another, and create a trail of information that moves through the organisation.

Putting an AI tool on top of that without fixing the underlying connections only gets you so far.

The real opportunity is to have the right information available, let systems work together and give AI a meaningful role inside those workflows. Sometimes that might mean handling a routine task. Sometimes it might mean finding information. Sometimes it might mean flagging something for a person to look at.

That is the direction Whatnxt is building towards: an HR ecosystem where businesses can start with what they need, connect applications as those needs grow and bring AI into the workflow where it can genuinely make a difference.

Because the goal is not to automate everything just because we can.

It is to take a closer look at the work itself and figure out where people add the most value, where technology can take some of the load, and where the two work better together.

That may be the real skill AI is asking us to develop not doing more work but getting better at deciding what needs us.