There is a fear sitting underneath a lot of conversations about AI at work.
What happens to us if the technology keeps getting better?
If AI can write the email, summarise the meeting, analyse the data, create the first draft, answer the customer and automate the admin, where does that leave the person?
It is not an irrational fear. Some tasks will disappear. Some roles will change significantly. Some businesses will use AI to reduce headcount. Pretending otherwise does not help anybody.
But there is another way to look at the same technology.
The question is not only what AI can take away from people.
It is also what AI can give back.
Time. Headspace. Confidence. Access to knowledge. The chance to spend less of the day on mechanical work and more of it on the things that still need a human being.
Judgement. Empathy. Context. Curiosity. Trust. Responsibility.
Used well, AI should not remove the human from work. It should create more room for the human part of work.
The fear is real, but the framing matters
A lot of AI discussion starts with replacement.
How many jobs can be automated? How many hours can be saved? How much work can one person now do with an AI assistant?
Those questions matter commercially. Businesses have to think about productivity and cost.
But if replacement becomes the only measure of success, we miss something important.
Most jobs are not one single task. They are a bundle of very different activities.
A finance role might include copying data between systems, reformatting reports, investigating anomalies, explaining results to management and challenging a decision that does not look right.
A customer-service role might include searching for information, typing routine responses, calming an upset customer and knowing when the normal process is not good enough.
A manager might spend time preparing notes, chasing updates, reviewing work, coaching people and making difficult calls when the answer is not obvious.
AI may be very good at some of those activities and much weaker at others.
That distinction matters.
The opportunity is not necessarily to automate the whole job. It is to remove the parts that do not need a person’s judgement so that more of their time can go into the parts that do.
The human in the loop is not a weakness
There is sometimes an assumption that the ultimate version of automation is a process with no human involvement at all.
For a lot of business work, that is not true.
Keeping a human in the loop is often exactly what makes AI useful rather than risky.
AI can produce a first draft quickly. A person can decide whether it is appropriate.
AI can summarise a long document. A person can recognise what matters in the wider context.
AI can identify patterns in data. A person can decide whether those patterns make sense and what action should follow.
AI can draft a response to a customer. A person can recognise that the customer does not need a technically correct answer. They need reassurance, flexibility or empathy.
That is not a failure of automation. It is good work design.
Research involving 758 Boston Consulting Group consultants helps explain why. In work led by Harvard Business School researchers, consultants using GPT-4 performed significantly better on tasks that sat within the technology’s capabilities. They completed work more than 25% faster and their human-rated performance improved by more than 40%. The researchers described this as a jagged technological frontier: AI can be extremely capable at one task and unexpectedly poor at another that looks very similar.
So the skill is not simply knowing how to use AI.
It is knowing when to trust it, when to challenge it and when the decision should stay with a person.
Better output often comes from a partnership
One of the most useful ways to work with AI is not to ask it to replace thinking, but to help improve it.
Instead of asking AI to make the decision, ask it to challenge yours.
- What have I missed?
- What assumptions am I making?
- What would someone who disagreed with this say?
- What evidence would weaken this conclusion?
- Can you explain this in another way?
Used like that, AI becomes less of an answer machine and more of a second pair of eyes.
That can improve quality because the human remains responsible for the outcome while gaining another perspective, a faster first draft or a structured way into a problem.
There is also evidence that AI can help less experienced people become more capable.
Researchers Erik Brynjolfsson, Danielle Li and Lindsey Raymond studied 5,179 customer-support agents after the introduction of a generative AI assistant. Productivity increased by 14% on average, but novice and lower-skilled workers improved by 34%. The researchers found suggestive evidence that the AI was helping spread some of the practices associated with stronger workers. You can read the NBER paper here.
That is important because it points to augmentation rather than simple substitution.
A newer employee does not have to wait years to encounter every useful pattern for the first time. AI can help surface relevant knowledge while the work is happening, with the person still there to apply it.
What we should automate first
If you are deciding where AI belongs in a business, start with the work that drains time without requiring much human judgement.
- Chasing information through inboxes.
- Moving data between systems.
- Reformatting the same report every month.
- Drafting routine summaries.
- Sorting, categorising and organising information.
- Searching through documents for something you know is in there somewhere.
- Creating the first version of something that a person will review anyway.
These are not unimportant tasks. They still need to happen.
But they are often poor uses of human attention.
And attention is one of the things businesses rarely account for properly.
When somebody spends half an hour copying information from one place to another, the cost is not only half an hour of salary. It is also half an hour they could not spend noticing a problem, helping a colleague, understanding a customer or thinking properly about a decision.
That is where automation can create something more valuable than speed.
It can create capacity.
What should we do with the time we get back?
Saving time is not the same as improving work.
If AI saves an employee an hour and the only response is to fill that hour with another hour of low-value output, we have made the process faster without necessarily making the job better.
The better question is: what can the human now do that the system could not?
They can have the conversation that needs sensitivity rather than sending the fastest possible response.
They can look at a set of numbers and ask whether the story behind them makes sense.
They can coach the new employee instead of just correcting the mistake.
They can spend longer understanding why a customer is unhappy rather than simply closing the ticket.
They can question an assumption, think through a consequence or notice when the standard process is producing the wrong result.
They can use empathy and judgement.
Those things are difficult to turn into neat productivity metrics, but they are often where the real value of a person sits.
There is a danger in handing over too much
A human-in-the-loop approach is not only about preserving jobs. It is also about preserving capability.
Microsoft Research surveyed 319 knowledge workers about their use of generative AI. Higher confidence in AI was associated with less reported critical-thinking effort. The researchers also found that the nature of critical thinking can shift towards verification, integration and oversight when AI is used. The full research is available here.
That can be useful, but only if the verification actually happens.
AI output is often polished enough to feel finished. That makes it very easy to stop questioning it.
If AI writes the analysis, proposes the decision and explains why the decision is right, a person can move from reviewing the answer to simply receiving it without really noticing the change.
That is where convenience can become dependency.
So there are some skills we should be careful not to automate out of ourselves.
We still need to know what good looks like.
We still need enough understanding to spot when something is wrong.
We still need to ask whether the answer fits the situation rather than simply whether it sounds convincing.
The better AI becomes, the more important those skills become.
Some jobs will still change
None of this means there will be no disruption.
The World Economic Forum’s Future of Jobs Report 2025 points to significant change in the labour market by 2030 and reports that many employers expect to reduce workforce numbers where AI can automate tasks. The report is available here.
That is uncomfortable, but it is part of the reality businesses and workers need to prepare for.
At the same time, employers continue to place value on capabilities such as analytical thinking, creative thinking, resilience, curiosity and leadership alongside technical skills.
That combination makes sense.
The future is unlikely to be purely human or purely automated.
It will be a question of how well we divide the work.
A simple way to think about it
Before handing a task to AI, there are three useful questions.
Does this task genuinely need human judgement?
If it does not, it may be a good candidate for automation.
Would AI make the human better at the task rather than simply remove them from it?
If AI can provide a first draft, another perspective, faster access to information or a useful challenge, the combination may produce better work than either would alone.
What will the person do with the capacity we create?
This one is easy to overlook.
The point of saving time should not always be to squeeze more output into the same day.
Sometimes the return on automation is having enough time to think properly.
AI should give us more room to be good at being human
People do not need another argument telling them not to worry about AI.
There are things worth worrying about.
Jobs will change. Poorly designed automation can remove useful skills. Businesses can use AI to make work more impersonal rather than less. And a confident AI answer is not the same thing as a good one.
But fear should not stop us seeing the other possibility.
AI can take some of the repetitive, mechanical work that fills our days.
It can help people reach knowledge faster and give less experienced employees a stronger starting point.
It can challenge our thinking, help us structure ideas and give us another perspective.
And when we keep people in the loop, it can improve the quality of the final result because somebody still brings context, judgement and accountability to the work.
Most importantly, it can give us more time for the things we should not want to automate away.
Empathy.
Judgement.
Curiosity.
Trust.
The difficult conversation.
The decision where there is no perfect answer.
Understanding what another person actually needs.
That is the version of AI businesses should be aiming for.
Not technology that makes humans less relevant.
Technology that takes more of the mechanical work off our hands, while giving us more room to do the human part properly.
