Companies often describe generative artificial intelligence as a productivity tool.
It can draft reports, summarise meetings, organise information and help employees complete work more quickly. This makes AI adoption appear to be primarily a technology decision: choose a platform, establish rules and train employees to use it.
The deeper organisational change may be happening somewhere less visible.
It is appearing in the skills companies no longer request when they advertise jobs.
New research suggests that businesses more exposed to generative AI have already reduced hiring demand for several activities traditionally associated with management and coordination. These include dividing work into tasks, sharing information, monitoring performance and handling routine operational problems.
Human judgement, however, has proved more difficult to remove.
The job advert reveals the organisation
Researchers Matteo Devigili, Erdem Dogukan Yilmaz, Vibha Gaba and Henrich Greve analysed recruitment patterns at 1,820 publicly listed companies in the United States.
The study, published in Management Science, examined job advertisements during the 12 months before and after ChatGPT became publicly available. The researchers compared companies whose occupations were more exposed to generative AI with firms facing lower exposure.
Their interest was not limited to whether companies hired fewer employees.
They wanted to understand whether firms were changing the kinds of organisational skills they valued.
The researchers call this skill deprioritisation: a systematic reduction in demand for human skills that organisations believe generative AI can perform or substantially support.
This matters because hiring decisions reveal more than recruitment needs. They offer an early indication of how companies expect work to be organised in future.
Coordination and monitoring lost ground
The largest changes appeared in structured activities that involve organising information or overseeing predictable processes.
Demand for skills related to dividing and sequencing work declined by about 23%. Requirements involving coordination fell by 21.3%, while demand for synchronous and asynchronous information-sharing skills declined by 18.5% and 21.4% respectively.
Monitoring skills fell by 23.2%.
Companies also advertised less frequently for skills associated with resolving routine operational exceptions, where demand declined by approximately 19%.
The pattern is understandable.
Generative AI can turn broad instructions into task lists, consolidate information from several documents, produce status summaries and identify recurring issues. When these activities require less human time, companies may conclude that they need fewer employees whose main value lies in performing them manually.
The effects also became stronger after the release of GPT-4, suggesting that companies responded not only to the publicity surrounding ChatGPT, but to improvements in what the technology could actually do.
Some decisions remained stubbornly human
Not every management skill declined.
The researchers found no statistically meaningful reduction in demand for skills involving staffing, matching people to work, designing incentives or resolving conflict.
These activities require more than processing information.
Allocating responsibility involves understanding who possesses the right experience, how personalities interact and which employee can exercise judgement when instructions become incomplete. Conflict resolution requires interpretation, trust and an understanding of relationships that may never appear fully in a dataset.
AI can provide information to support these decisions.
The findings suggest that companies remain less willing to hand over the decisions themselves.
This creates a more precise view of workplace automation. Generative AI may not remove management as a function. It may instead change what managers are expected to contribute.
Routine coordination becomes less valuable. Judgement becomes more valuable.
The risk is organisational memory
Reducing unnecessary work can improve efficiency. It can also create a vulnerability.
Employees learn partly by performing the routine tasks that organisations are now beginning to automate. A junior employee who prepares reports, coordinates projects or investigates operational exceptions gradually learns how the organisation functions.
When those responsibilities disappear, the work may be completed faster, but the company may also weaken the route through which future expertise develops.
The researchers describe the possibility of organisational knowledge becoming hollowed out. A company may retain the final decision-makers while gradually losing the people who understand how information was gathered, why processes were designed in a particular way and when an automated recommendation should be challenged.
The immediate efficiency gain may therefore create a longer-term capability problem.
What this means for South African employers
For South African organisations, the practical question is not whether every declining skill should be preserved.
It is whether the business understands which skills are genuinely redundant and which are still necessary for learning, oversight and succession.
A bank may use AI to summarise policies and draft internal reports. A retailer may automate performance monitoring. A professional-services firm may allow employees to generate project plans within minutes.
These changes can release valuable capacity.
But an organisation that stops developing employees who understand the underlying work may become increasingly dependent on systems it cannot properly evaluate. That risk is particularly important where specialised institutional knowledge takes years to build and is concentrated among a small number of experienced employees.
Companies should therefore distinguish between automating a task and abandoning the capability behind it.
The findings remain an early signal
The study does not prove that companies have already completed large-scale internal restructuring.
It examined job advertisements from publicly listed American companies during the early period following ChatGPT’s release. Job adverts reflect hiring intentions, not necessarily the full responsibilities employees eventually perform.
The researchers also caution that large listed companies may respond differently from smaller businesses, privately held firms and organisations operating under different labour or regulatory conditions.
The findings should therefore be read as evidence of changing expectations rather than a final picture of the AI-shaped company.
Even so, recruitment is where organisational strategy often becomes visible.
Companies may still speak publicly about reskilling employees and creating human–AI collaboration. Their job advertisements reveal which human capabilities they are already beginning to value less.
Generative AI is therefore becoming more than a tool employees use.
It is becoming a force that changes how companies divide work, where they place responsibility and what they believe managers are still needed to do.
The strategic question is no longer simply how much work AI can complete.
It is whether organisations can remove human effort without also removing the knowledge and judgement they will eventually need.
Source Information
Study Title: Skill Deprioritization: Reorganizing in the Age of Generative Artificial Intelligence
Authors: Matteo Devigili, Erdem Dogukan Yilmaz, Vibha Gaba and Henrich Greve
Journal: Management Science
Published online: 24 July 2026
DOI: 10.1287/mnsc.2025.01859
