Beyond the headline: what AI is really replacing at work
Every week, another headline announces jobs being eliminated by AI.
But there is an important distinction that often gets lost in the story:
AI rarely eliminates an entire job. It eliminates specific layers of work within a job.
That distinction matters — especially for managers making decisions about headcount, restructuring, and the future of their teams.
A documented case: Salesforce
Marc Benioff, Salesforce’s CEO, has publicly stated that the company reduced its customer support organization from around 9,000 employees to approximately 5,000 after deploying AI agents.
At first glance, this looks like a straightforward example of AI replacing people.
But the more useful question is:
What work was actually replaced?
AI agents can increasingly handle parts of customer support such as initial triage, classification, routing, and routine first responses.
That does not mean the entire support role has disappeared.
It means that one layer of the role — the layer with highly structured inputs and predictable outputs — has become increasingly automatable.
That distinction is easy to miss when the headline simply says: AI cut thousands of jobs.
The data is more complicated than the headlines
The broader evidence also challenges the simple story of AI arrives → jobs disappear.
A Harvard Business Review analysis has argued that many layoffs publicly attributed to AI are difficult to separate from broader cost-cutting and investor-signaling strategies.
One particularly striking finding is that roughly 92% of companies announcing AI-related layoffs increased their overall headcount during the same period.
In other words, companies may be shrinking certain functions while expanding others.
Gallup data adds another counterintuitive signal: among workers who were laid off, only a small fraction reported AI itself as the primary reason for the job loss. Employees who did not use AI were, in some cases, more exposed to layoffs than employees whose work was directly affected by AI.
The picture is therefore more complicated than AI replaces humans.
The real story is often about which tasks are changing, which capabilities are becoming less valuable, and where human work is being redeployed.
So what actually gets eliminated?
The highest-risk layer of a job tends to have three characteristics.
1. The input and output are clearly defined
The task follows a recognizable pattern.
For example:
- classifying a support ticket
- extracting information from a standard document
- screening a résumé against predefined criteria
- producing a routine summary
2. The task requires little judgment or accountability
The consequences of a wrong answer are relatively limited.
There is little ambiguity about what should happen next, and someone else can take responsibility for the final decision.
3. The value does not depend on human interaction
Some work is valuable not simply because a problem gets technically solved, but because of how the interaction happens.
- Consider an angry customer.
- Resolving the technical issue is one task.
- Understanding the customer’s frustration, deciding how much flexibility to offer, protecting the relationship, and taking responsibility for the outcome are different layers of work.
The first may be highly automatable. The second may not be.
The job is not the unit of automation
This is where many AI-driven workforce decisions go wrong. A job title is a bundle of different tasks.
- Some are repetitive.
- Some require judgment.
- Some require context.
- Some require accountability.
- Some depend on trust and human interaction.
AI does not necessarily attack the entire bundle at once. It tends to attack the most structured and predictable layer first.
That means the right question is not:
Can AI do this job?
It is:
Which parts of this job can AI perform reliably — and which parts should remain human?
That is a very different management question.
Why this distinction matters for managers
Eliminating an entire role because part of it can be automated can create two problems.
First, the organization may remove the human layer that was actually creating value. Second, the organization may mistake cost reduction for productivity improvement.
If a company removes ten positions but also removes the judgment, quality control, customer relationship, or accountability those people provided, the financial saving may be accompanied by a hidden operational cost.
The question is therefore not simply how many people AI can replace.
It is:
What value disappears when those people disappear?
The HENARCO view
The strategic decision is not about protecting every existing job from automation. Nor is it about replacing people simply because a technology can perform part of their work.
The better approach is to deconstruct the role. Identify the tasks AI can perform reliably. Identify the tasks that require judgment. Identify the tasks where accountability matters. And identify the interactions where the human relationship is itself part of the value. Then automate deliberately.
Not the job. The right layer of the job.
Better Questions
Before eliminating or significantly shrinking a role because of AI, a manager should ask:
- Do we have measured evidence that AI can perform this work reliably, or are we reacting to the AI narrative?
- Which parts of the role require judgment, context, or accountability — and are we accidentally removing those parts as well?
- Where does human interaction create value beyond simply completing the task?
- If we remove this role, what quality, trust, judgment, or accountability might disappear with it?
- Are we actually improving productivity — or simply reducing headcount and calling it AI transformation?
The future of work may not be defined by how many jobs AI eliminates.
It may be defined by how intelligently organizations redesign the jobs that remain.