What Is Happening: Work Is Being Broken Into Automatable Tasks
The public debate normally asks whether AI will eliminate a particular job: accountant, engineer, designer, doctor, lawyer, customer-service representative or manager. That framing is too crude. Organisations do not automate job titles. They automate tasks, decisions and handoffs inside jobs.
An accountant’s role contains data entry, document reconciliation, regulatory interpretation, client communication, exception handling and professional accountability. Some of those activities are highly structured and automatable. Others depend on context, trust and judgment. The same applies to engineering, healthcare, law, marketing and management.
The International Labour Organization’s refined 2025 index found that approximately one in four jobs worldwide falls within an occupation potentially exposed to generative AI. Yet its central conclusion was that transformation, rather than complete replacement, remains the more likely outcome. Clerical work carries the greatest exposure, while increasingly capable systems are also affecting digitised professional work in media, finance and software.
That finding is far more credible than sensational forecasts claiming a fixed percentage of humanity will simply become redundant. Exposure is not the same as elimination. A job may be exposed because AI can perform 20 percent, 50 percent or even 80 percent of its tasks, but the remaining work may still require a qualified person, especially where failure carries financial, legal, medical, engineering or reputational consequences.
The original workplace-AI argument correctly recognised that augmentation could develop into replacement in certain functions and that repetitive back-office processing would face early pressure. It also raised the essential concern that workers may stop developing professional judgment if “the robot told us” becomes an acceptable explanation for a decision.
That concern has aged well. The deterministic numbers have not.
What It Actually Means: The First Draft Is Becoming Almost Free
A large share of office work is not difficult because the final decision is extraordinary. It is difficult because reaching a usable starting point consumes time.
Someone must open the files, locate the relevant paragraphs, sort important information from noise, compare previous records, identify missing items and prepare an initial response. Generative AI compresses this early stage. The employee no longer begins with a blank screen or an unorganised pile of material; the employee begins with a draft, summary, classification or proposed sequence of actions.
This is where measurable productivity gains appear. An OECD review of experimental evidence found that generative AI produced average gains ranging from roughly 5 percent to more than 25 percent in activities involving customer support, software development and consulting. The same evidence also showed that performance gains depend heavily on the task, the worker’s ability and the worker’s capacity to evaluate the output. When AI is applied beyond its capabilities, performance can deteriorate because the system introduces errors or lowers quality.
This creates a workplace contradiction. AI can make work faster while making management more demanding. Faster drafting means more output, but more output requires more verification. Employees may complete individual tasks sooner while receiving a greater volume of tasks overall.
A 2026 World Economic Forum report on entry-level work captured this tension sharply: 68 percent of surveyed entry-level workers reported productivity improvements from AI, while 45 percent said AI had also caused them to spend more time working overall.
Efficiency, therefore, does not automatically become leisure, better pay or better work. Without deliberate management, it merely raises expectations.
What Nobody Is Telling Employees: AI Can Remove the Training Ground Beneath Senior Expertise
Every senior professional once performed junior work.
An engineer learns by checking drawings, calculating loads, inspecting installations and discovering why assumptions fail on site. A lawyer develops judgment by reading cases and drafting documents. A financial analyst learns by building models, not merely approving them. A manager learns operations by handling routine problems before being trusted with exceptional ones.
When organisations automate entry-level work too aggressively, they may save money today while destroying the apprenticeship pipeline that produces tomorrow’s experts.
This is one of the least discussed consequences of workplace AI. The junior employee may appear inefficient compared with a machine-assisted senior professional, but that junior role is also a learning mechanism. Remove the work through which knowledge is acquired, and the organisation eventually discovers that it has plenty of AI-generated output but too few people capable of judging whether it is correct.
The World Economic Forum’s broader 2025 employment analysis projected substantial churn by 2030: 170 million roles created, 92 million displaced and a net increase of 78 million. It also estimated that nearly 40 percent of workplace skills would change and that 59 out of every 100 workers would require training, reskilling or upskilling.
The numbers do not support complacency. They also do not support fatalism. They support preparation.
